Energy saving policies in grid computing and smart environments

UNIVERSIDAD DE SEVILLA DEPARTAMENTO DE LENGUAJES Y SISTEMAS INFORMÁTICOS Energy-Saving Policies in Grid-Computing and Smart Environments International Doctoral Dissertation presented by D. Alejandro Fernández-Montes González advised by Dr. Juan Antonio Ortega Ramírez and Dr. Luis González Abril . July 2012. A mis padres. Agradecimientos Quisiera expresar mi más sincero agradecimiento a mis directores de Tesis, Dr. D. Juan Antonio Ortega Ramírez y Dr. D. Luis González Abril, por todo el esfuerzo y tiempo dedicado a la elaboración de este trabajo. Han conseguido encontrar el equilibrio perfecto entre el apoyo y la exigencia, necesarios para abordar esta tarea. Del mismo modo, al Dr. D. Francisco Velasco Morente por los buenos consejos que me ha dado y la sabiduría que siempre comparte con los demás. A mis padres Fernando y M ª Victoria por haberme dado las facilidades y el apoyo necesario para realizar los estudios universitarios y a mis hermanas Olga y Lucía por su cariño y ánimos. A todos mis compañeros del departamento de Lenguajes y Sistemas Informáticos, que hemos compartido este largo viaje y con los que siempre ha habido un compañerismo ejemplar. En especial quiero agradecer a Cristina, Diana, Fermín, Javi, Irene, Guti, Bea, Juan Antonio, Pablo, María, Ángel, Luisa y Mayte por los buenos ratos que hemos pasado dentro y fuera de la escuela. A mis compañeros de despacho Juan Antonio e Isabel por aguantarme y compartir tantas horas y a Adela por contagiarme su simpatía durante los desayunos. A los miembros del grupo ENTI de la Universidad Carlos III que me acogieron cuando empezaba la carrera investigadora, y a los compañeros del Laboratoire de l'Informatique du Parallélisme del ENS de Lyon por darme las bases para el desarrollo de esta tesis. A todos, muchas gracias. CONTENTS Abstract xiii Resumen xv I Preface 1 Introduction 19 1.1 ResearchMotivation ........................... 19 1.2 Research Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . 22 1.3 ResearchQuestion ............................ 23 1.4 SuccessCriteria.............................. 23 1.5 Denition of Energy eciency . . . . . . . . . . . . . . . . . . . . . . 24 1.5.1 Energy versus power . . . . . . . . . . . . . . . . . . . . . . . 24 1.5.2 Ratio of inputs to outputs . . . . . . . . . . . . . . . . . . . . 24 1.6 Environments analyzed and discussed . . . . . . . . . . . . . . . . . . 25 1.7 Thesisoutline............................... 25 2 Energy-Saving in Grid-Computing Environments 29 Page vii CONTENTS 2.1 Introduction................................ 30 2.2 Grid-Computing Motivation . . . . . . . . . . . . . . . . . . . . . . . 31 2.3 Power Management Layers . . . . . . . . . . . . . . . . . . . . . . . . 32 2.3.1 Why power management . . . . . . . . . . . . . . . . . . . . . 32 2.3.2 Layers of the Server Power Management . . . . . . . . . . . . 36 2.3.3 Power Management Policies . . . . . . . . . . . . . . . . . . . 39 2.4 Grid'5000 at a glance . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 2.4.1 Jobs ................................ 41 2.4.2 Resources ............................. 42 2.5 Graphical representation . . . . . . . . . . . . . . . . . . . . . . . . . 44 2.6 Data envelopment analysis . . . . . . . . . . . . . . . . . . . . . . . . 46 3 Energy-Saving in Smart Environments 49 3.1 Introduction................................ 50 3.2 WSN to sense Smart Environments . . . . . . . . . . . . . . . . . . . 51 II Selected research works 4 Smart scheduling for saving energy in Grid computing 57 5 Evaluating decision-making performance in a Grid-computing environment using DEA 69 6 A Study on Saving Energy in Articial Lighting by Making Smart Use of Wireless Sensor Networks and Actuators 83 Page viii CONTENTS III Other relevant research works 7 Pervasive Computing Approaches to Environmental Sustainability - Automating Smart Environment Systems 93 8 Smart environment vectorization - An approach to learning of user lighting preferences 101 9 Modeling Smart Homes for Prediction Algorithms 113 10 Smart Environment Software Reference Architecture 125 11 LECOMP: Low Energy COnsumption Mesh Protocol in WSN 135 IV Final remarks 12 Conclusions and future work 149 12.1Conclusions ................................149 12.2Futurework................................150 A Curriculum 153 A.1 Publishedpapers .............................153 A.1.1 JCR Indexed Journals . . . . . . . . . . . . . . . . . . . . . . 154 A.1.2 International Conferences . . . . . . . . . . . . . . . . . . . . 155 A.1.3 Bookchapters...........................158 A.1.4 OtherJournals ..........................158 A.1.5 National Conferences . . . . . . . . . . . . . . . . . . . . . . . 158 Page ix PART I Preface Page 17 CHAPTER 1 INTRODUCTION E=m·c2 Albert Einstein, 1905 1.1 Research Motivation Energy eciency and sustainability are two areas of great interest in recent decades. CO2 emissions worldwide continue growing signicantly as shown in Figure 1.1 and detailed in Table 1.1. From 1990 to 2009 CO2 emissions rose by approximately 38% while population grew about 28% [60], which indicates 10 points of growth in percentage terms. There are some good news: energy generation is now more ecient as can be observed in Table 1.1. The kilograms of CO2 per Gross Domestic Product (GDP) is currently about 15% less than in 1990, which means that the Page 19 Chapter 1: Introduction generation of products and services around the world is also more ecient. However the eorts of both developed and developing countries, and the commitment for a more sustainable world cannot be relaxed. Developing countries, such as China and Brazil, have greatly increased their energy requirements over the last decade, and have used non-sustainable sources of energy, such as coal and oil. 21536.2 22110.9 23763.5 27708.5 29354.8 29967.1 29549.3 8782.3 9241.1 10031.9 11466.6 12053 12273.7 12149.8 24257.5 27196 32174.3 36896.6 39885.3 40470.4 39674.4 5266.9 5680.5 6075.5 6455.4 6607.2 6684 6760.7 1990 1995 2000 2005 2007 2008 2009 CO2 Reference Approach (Mt of CO2) TPES (Mtoe) GDP (billion 2000 USD) Population (millions) Figure 1.1: Key Indicators of World CO2 Emissions (1) The Reference Approach and the Sectoral Approach often give diering results since the Reference Approach is a top-down approach using a country's energy supply data and has no detailed information on how the individual fuels are used in each sector. The Reference Approach provides estimates of CO2 for comparison (1) TPES : Total Primary Energy Supply. PJ : Peta-Joule ( 1015 Joules). MTOE : Million Tons of Oil Equivalent. GDP : Gross Domestic Product. 2000-USD : The value of a United States Dollar in year 2000. The reference price taken. PPP : Purchasing power parities are the rates of currency conversion that equalise the purchasing power of dierent currencies. A given sum of money, when converted into dierent currencies at the PPP rates, buys the same basket of goods and services in all countries. In other words, PPPs are the rates of currency conversion which eliminate the dierences in price levels between dierent countries. Page 20 1.1 Research Motivation 1990 1995 2000 2005 2007 2008 2009 Var. Rate 90-09 CO2 20966 21791 23492 27188 29047 29454 28999 38.30%Sect. Approach (Mt) CO2 21536 22110 23763 27708 29354 29967 29549 37.20%Ref. Approach (Mt) TPES 367696 386906 420014 480084 504633 513874 508690 38.3% (PJ) TPES 8782 9241 10031 11466 12053 12273 12149 38.3% (Mtoe) GDP 24257 27196 32174 36896 39885 40470 39674 63.6%(billion 2000 USD) GDP PPP 33340 37834 45799 55547 62111 64095 64244 92.7%(billion 2000 USD) Population 5266 5680 6075 6455 6607 6684 6760 28.4% (millions) CO2 /TPES 57 56.3 55.9 56.6 57.6 57.3 57 0.0% (t CO2 /TJ) CO2 /GDP 0.86 0.8 0.73 0.74 0.73 0.73 0.73 -15.4%(kg CO2 ) /2000 USD) CO2 /GDP PPP 0.63 0.58 0.51 0.49 0.47 0.46 0.45 -28.2%(kg CO2 / 2000 USD) CO2 /population 3.98 3.84 3.87 4.21 4.4 4.41 4.29 7.8% (t CO2 /capita) Table 1.1: Worlwide CO2 emissions indicators. Page 21 Chapter 1: Introduction with estimates derived using a Sectoral Approach. Theoretically, it indicates an upper bound to the Sectoral Approach: "1A fuel combustion", because some of the carbon in the fuel is not combusted but instead emitted as fugitive emissions (as leakage or evaporation in the production and/or transformation stage). Within this context the increasing energy consumption of Information Technology (IT) companies and infrastructures propel us towards nding solutions which enable us to save energy while maintaining or even improving current services. Information Technology energy consumption alone represents a monstruos 3% to 5% of CO2 emissions worldwide which is similar to that of aviation transport, and hence it is worth investing time and resources in order to analyze and propound methods to save energy within IT. Moreover, buildings in Spain currently cover a sum total of 3,500 million square meters, and consume 17% of the total energy consumption [46] and therefore it seems that research into this topic is also worthwhile. 1.2 Research Methodology This research follows the standard scientic research technique [110] which includes the following phases: 1. Literature Review. The research literature phase formed the initial stage of the research question through the retrieval of information, tools and methodologies, and consequently via the study and observation of these fundamentals. 2. Modelling. During the modelling phase the environments expressed in the research question were analyzed and studied. This model includes the computational artifacts that model the Grid design. 3. Algorithmic Development. An innovative technique with its algorithm have been developed in order to handle and cope with the various issues involved. Other techniques were developed in order to enable rigorous comparisons. Page 22 1.3 Research Question 4. Prototyping and Evaluation. The simulation tools based on the model have been built in order to run the experiments and to test and evaluate the results. 1.3 Research Question The research question that leads this thesis dissertation is: Which are the best policies to save energy in smart environments and in Grid-Computing environments? Within the context of Grid-Computing, we focused on saving energy through deploying and undeploying resources in a Grid environment in order to save the most energy. To this end, energy-saving policies from the literature and new proposals have been tested and compared whilst making use of simulation software specially developed for this purpose. Within the context of Smart Environments, we focused on saving energy through analyzing user preferences in lighting conditions, and by adapting the environments to these conditions. The question becomes how to adjust lighting conditions in order to save the most energy whilst simultaneously satisfying user preferences. 1.4 Success Criteria Success will be achieved if the research question is resolved. This means that we have to check that both the model and the supporting algorithms indeed dene energy-saving policies and demonstrate their validity by the experiments through the simulation software. The simulation software demonstrates that the output results match the results estimated by manual calculation. Within the Smart Environment Infrastructure, lighting conditions over a period of months, and an estimation of results on saving energy in lighting equipment were computed for this period. Within the Grid-Computing context, the experiments performed on the simu- Page 23 Chapter 1: Introduction lation software were carried out accross eight locations of the Grid infrastructure, whereby each location was composed of a set of clusters, and each cluster was compounded of a set of machines/resources. 1.5 Denition of Energy eciency The term energy eciency is currently bandied about in politics, board meetings, research and in the press. Since energy eciency constitutes the focus of this thesis, this term needs reviewing, and it can give an insight into the work already carried out in this eld. 1.5.1 Energy versus power Energy is a scalar physical quantity dened as the ability to exert pulls or pushes against the basic forces of nature along a path of a certain length, or equally, the capacity to do work ( W ). Thus, energy is usually reported in kilowatt hours, (kWh) while taking into account the following equation: kWh = (Watts ×tuse)/1000 . Energy is a combination of power and time. The fewer watts or the shorter duration, the more energy saving. In the International System of Units (SI), energy is measured in joules ( J ) which is the energy expended in applying a force of one newton through a distance of one metre, or in passing an electric current of one ampere through a resistance of one ohm for one second, that is, J= (kg ·m2)/s2=N·m=Pa ·m3=W·s 1.5.2 Ratio of inputs to outputs Secondly, it is important to recognise that eciency is dened as the ratio of inputs to outputs. If the outputs increase faster than the inputs, then eciency is improving. In terms of clusters, grids, data center, and IT infrastructures in general, energy Page 24 1.6 Environments analyzed and discussed inputs are in kWh and outputs are some degree of operation of the intended IT hardware. In terms of lighting, energy inputs can be considered in kWh and outputs as the quantity of light perceived by humans. The goal is to adjust this power in order to satisfy human preferences about lighting conditions. 1.6 Environments analyzed and discussed Energy saving and energy eciency is approached from two sides: Grid-computing environments and smart environments. In relation to Grid-Computing, we tackled energy saving in collaboration with the Laboratoire de l'Informatique du Parallélisme of the Ecole Normale Supérieure de Lyon, where permanent researcher Laurent Lefèvre leads the research team on saving energy in Grid-Computing and on other areas. The Laboratoire de l'Informatique du Parallélisme has 125 members and is one of the most active in research and with regard to educational topics and tasks. In relation to Smart Environments, this research topic has been the main focus of the research group Idinfor (to which the Ph.D candidate belongs) ever since it was formed in 2003. The work presented in this document provides a solution for the improvement of lighting conditions in smart environments, by adjusting infrastructures in order to satisfy user preferences, by and minimizing energy consumption. 1.7 Thesis outline The document is structured as follows: Chapter 2 introduces the problem of saving energy in IT infrastructures such as Grids and Data Centers. Chapter 3 presents the domain in which saving energy in Smart Environments is developed. In Part. II, three selected journal papers are provided. These journals are included in the Page 25 Chapter 2: Energy-Saving in Grid-Computing Environments mization of the impact on the availability and Quality of Service (QoS) of the Grid environment. To this end, several energy-saving policies are proposed and compared in the highest layer (see the following Section 2.3) the so-called Data Center layer. 2.3 Power Management Layers 2.3.1 Why power management Power management enables the system administrator to save energy and to cut costs by using available tools. Server power management is the principal weapon in the ght against rising energy needs and costs. These capabilities can make a great dierence in the energy consumption of huge computing infrastructures, such as Grids and data centers. Eciency starts with operations, exactly how the operations manage the services versus how the power is allocated. Eective power management enables us to use less power in the data center, thereby cutting energy costs. This can be observed in one example from a recent performance comparison in Figure 2.1. The top line shows the power consumption of an older server with an older operating system and no power management. The bottom line shows the power consumption of a newer server with a new operating system and with power management enabled. The Y-axis represents power consumption, and the X-axis represents performance. As can be observed, the server with active power management consumed less power: that is, a 30 percent drop at 10 percent utilization, and a 14 percent drop at 100 percent utilization. The dierence in throughput is also striking. The modern server with power management engaged demonstrates far better throughput at all levels. There is also a marked improvement in performance. The newer server with power management implemented delivers almost double the performance at less than 90 percent of the power used by the older server. Since servers sometimes sit idle, it is also worth noting that the newer server is much more energy ecient than the older Page 32 2.3 Power Management Layers server when operating at idle power settings. 152 164 185 201 218 240 251 72.6 116 126 134 144 156 169 181 194 207 218 0 50 100 150 200 250 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Power (Watts) Performance Comparison of Power Consumption w2k3 w2k8 Figure 2.1: Comparison of power consumption between servers. The benets of server power management: Eective power management leads to a waterfall of benets. Like dierent currents of water rushing together, the benets of server power management combine, building on each other and multiplying in intensity. Over time, they become strong enough to have a signicant impact. OPEX and CAPEX: Operating expenses or OPEX are the ongoing costs for running a business, a product or system. Its counterpart, capital expenditure or CAPEX is the cost of developing or providing non-consumable parts for the product or system. For example, the purchase of a printer involves CAPEX, while the paper, toner, and power costs are all OPEX. Eciency starts with operations, but the cutting of energy expenditure will also Page 33 Chapter 2: Energy-Saving in Grid-Computing Environments trigger another benet: long-term CAPEX. The less energy is consumed, the less heat is created and therefore the less demand there is for the cooling devices, which in turn involves less consumption of energy and water and increases the useful lifetime of hardware. Moreover, if the heat generated per server is reduced, then the number of servers per rack can be increased. Hence, when the Grid or the data center receives more demand for services, this feature enables the administrator to t more servers into the same space and probably prevents the unnecessary purchase of from purchasing new facilities Productivity: The more servers tted into each rack, and the more racks in each facility, the more productivity is obtained. Increasing the capacity enables the organization to be more productive and protable. Carbon Cost: A reduction in OPEX implies a reduction in CAPEX, an increase in productivity, and also has the potential to reduce carbon cost. If a data center consumes power generated from fossil fuels, the consumptiom of less power will reduce its carbon footprint. Reducing a facility's carbon footprint can help us address environmental concerns by decreasing our contribution to the greenhouse gas emissions created during fossil-fuel-based power generation. If carbon reduction credits are applicable, our company might greatly benet in terms of tax from a reduction in the consumption of power generated from fossil fuels. Depending on the source of the energy consumed, a greater or lesser quantity of CO2 is generated. In this thesis, it is assumed that the emissions of CO2 are of 0.274 kg per kWh which belongs to the production of electricity from oil [84]. Power Management Today: In this way, the waterfall keeps on owing. Reduced power consumption sets it all in motion. The benets then build upon each other, by multiplying their impacts, and transforming the data center and organization over time. It all adds up. Page 34 2.3 Power Management Layers Awareness of the tremendous potential of power management is growing, and its importance has been recognized by professionals in major corporations worldwide. Eective power management techniques have been developed, applied, and rened, and new techniques are constantly being developed. The tools of server power management are widespread throughout the data center world. What you might already be using: Many of these features are within reach of data centers right now. At the component level, manufacturers have improved their chips to perform power monitoring and have enabled them for integration into the functions of power management systems. Myths: Features of server power management are related to certain drawbacks, although many are simply myths.  Reliability. The reliability of power-saving features in microprocessors and other hardware devices has been proven through extensive testing and evaluation by vendors. Today, power management constitutes a mature technology and equipment is subjected to rigorous testing to ensure quality and robustness. In fact, vendors are so condent of these features and that they are usually enabled by default.  Performance. Low-level power management features are only activated when a device or component is idle, and when performance is required, these features are automatically switched o.  Availability. It is true that low-power states can aect availability, and highlevel power management can delay response time of services, and hence policies have to deal with this drawback, by waking-up systems during high-workload periods. Page 35 Chapter 2: Energy-Saving in Grid-Computing Environments 2.3.2 Layers of the Server Power Management Reducing OPEX, CAPEX, increasing productivity and decreasing carbon costs are tackled from ve basic layers of server power management. The basic ve layers are explained in the following subsections and are usually shown as a pyramid such as that in Figure 2.2. Component Physical Operating System Rack DataCenter Figure 2.2: Layers of the Server Power Management. Component Layer: Components are the individual pieces that compound a machine, including chipsets, processors, memory and I/O devices. These components are often manufactured with power management capabilities, such as frequency and voltage scaling. By combining frequencies and voltages, a processor enters into a performance state (P-state). P-states imply reduced performance. C-states are examples of idle states of processors as specied by the Advanced Conguration and Power Interface (ACPI). C-states range from C1 to C6, with C6 using the least power and having the most latency. Other examples of idle states include: Page 36 2.3 Power Management Layers  PPFE, a fast exit state that memory enters when idle  PPSE, slow exit idle memory state  Self-refresh, a low-power but long-latency idle memory state, and  ACPI's B1-B4, D1-D4 and PCIE's L1 to L4 states Physical System Layer: In the physical system layer, the whole system (i.e., the server) controls the power states of its components. In the ACPI, the terminology assigns a letter to each type of component (B-states refer to I/O buses, L-states to serial such as PCI and G-states, which describe the state of the system as a whole). For each of these types, the "0" state represents the state where the component is fully functional and no power-saving is achieved, and the subsequent states represent the increasing levels of energy saving and the increasing periods of response to return to full power and performance. Additional power management can be performed at this level by utilizing hardware capabilities, such as the use of blade servers. Blade servers share common resources within a blade chassis, and hence vendors incorporate power-capping technologies into these chassis, by limiting power per blade. The chassis manage the power of the individual blades, regardless of what operating systems the run. The chassis can also optimize power distribution according to the needs. Operating System Layer: This layer collaborates with the hardware in order to make best choices towards saving energy. This management can occur automatically, as some operating systems are set up in a power-saving mode by default. An example of this is the feature called Core parking, which consists of using only a few cores, thereby allowing other cores to sleep. Within this layer, the power capping also acts, which can be used to prevent systems from consuming more than a specied level of power. This feature articially limits power consumption of a server or set of servers according to external inputs or policies, and the operating system is in charge of the application of these limits. Page 37 Chapter 2: Energy-Saving in Grid-Computing Environments Rack Layer: This layer is in charge of managing power consumption of all equipment housed within a given rack. A rack usually includes blade chassis, rack-mount servers, and three-phase power whips. In order to save energy, the power used can be limited for the rack as a whole. Power management tools are needed, which manage limits at previous levels in order to meet the requirements of the rack level. These tools, usually known as Aggregation software, obtain and aggregate power consumption information from a group of servers and other devices. This software facilitates the management of the consumption of multiple servers, and queries the consumption information through the network, via out-of-band communication. The out-of-band data stream is seen as separate from the main, or in-band, data stream and carries instructions than can remotely power the systems. Servers must be equipped with both chip- and system-level power-saving features. Data Center Layer: This layer includes everything in the data center: all the racks, all the servers in those racks, and all the power-saving capabilities in all the hardware and software in the room. At this level, the key is to look at the whole room, monitoring the overall power consumption and using software to redistribute the workload to reduce power consumption and to control heat dissipation. For example, a manager could consolidate the load on a set of servers and switch other servers o. Another aspect at this level involves the control of the physical layout of the room, for example increasing the number of servers per rack. Focus of Power Management Layers: The layers range in increasing complexity from the component layer to the data center layer. The rst three (component, physical, and operating system) all have a single-system focus. The rack and data center layers are more complex, have a multi-system focus, and are more concerned with data collection and management. Activities performed in any layer can aect any other layer. Page 38 2.4 Grid'5000 at a glance 2.3.3 Power Management Policies Denition of Power Management Policy: Policies must meet the organization's needs and priorities and evaluate performance versus power. This thesis tackles energy-saving policies at the highest layer, that of the data center layer, by evaluating energy consumption for dierent policies whose main goal is to minimize energy consumption. Policy Examples: Policies at the data center level can be based on contracts with power providers that dictate a maximum level of consumption. The manager can establish a policy to ensure that power consumption remains below a specied level. Simple Policies: Policies can be categorized as either simple or complex. Simple policies are triggered by simple rules, for example, turning o a number of servers during late-night hours or periods of seasonal low activity. Complex Policies: These policies include more factors, and determine what will happen and when based on external factors. Such policies can respond dynamically to events as they happen. For example, a complex policy can respond to the amount of workload by activating or deactivating servers on demand. The work in this document is focused on these kind of complex policies, which are presented in the papers attached to Part II. 2.4 Grid'5000 at a glance Grid'5000 has been built upon a network of dedicated clusters. It is not an ad hoc Grid. The infrastructure of Grid'5000 is geographically distributed over various sites, of which the initial 9 are located in France: Bourdeaux, Grenoble, Lille, Lyon, Page 39 Chapter 2: Energy-Saving in Grid-Computing Environments Nancy, Orsay, Rennes, Sophia-Antipolis and Toulouse. Porto Alegre, in Brazil, and Luxemburg, are currently being ocialy included as the 10th and 11th sites respectively. Reims (France) has been recently added as the 12th site. The project began in 2004 as an initiative by the French Ministry of Education and Research, INRIA, CNRS, the Universities of all the aforementioned cities, and several regional councils. Figure 2.3 shows the geographical representation of the Grid'5000 nodes. Figure 2.3: Geographical distribution of Grid'5000 nodes. The initial aim was to achieve 5000 processors on the platform. This objective, adjusted to 5000 cores, was reached during the winter of 2008-2009. On 16th March 2010, 1569 nodes (5808 cores) were in production in Grid'5000. Today the Grid'5000 counts on 8580 cores from 2956 processors. These sites can now connect to each other within the same VLAN at 10Gbps thanks to the dark bre infrastructure, in an incomplete graph scheme. This network infrastructure is provided by Renater, the French National Telecommunication Network for Technology, Education and Research. It provides interconnection with GEANT-2, overseas territories, and the SFINX (Global Internet exchange). Page 40 2.4 Grid'5000 at a glance In terms of the software needed for the use, deployment and management of Grid'5000, the following tools must be mentioned:  OAR 2. OAR is a resource manager (or batch scheduler) for large clusters. It allows cluster users to submit or reserve nodes either in an interactive or a batch mode.  Kadeploy 3. Kadeploy is a fast and scalable deployment system for cluster and Grid-Computing. It provides a set of tools for cloning, conguring (post installation), and managing a set of nodes. Currently it successfully deploys linux, *BSD, Windows, and Solaris on x86 and 64-bit computers.  KaVLAN. VLAN is a manipulation tool for network isolation of experiments. Grid'5000 enables researh experiments to be carried out at Grid or at cluster level, and guarantees hardware and bandwidth of a more homogeneous nature, although Grid-level experiments are preferred in planning. Each site of Grid'5000 hosts several clusters, since hardware is acquired in incremental steps on each site, whereby clusters are formed at each purchase. Each cluster is composed of two kinds of nodes:  Compute nodes, which constitute the base elements of a cluster, upon which computations are run.  Service nodes, which are dedicated to hosting the Grid infrastructure services, such as control and deployment. Each node can supply several cores, which are the nest granularity of resource in Grid'5000. This means that if a machine has a microprocessor with n cores, it oers n resources to the Grid. 2.4.1 Jobs The platform can be used in two separate modes: submission or reservation. Page 41 CHAPTER 3 ENERGY-SAVING IN SMART ENVIRONMENTS E=~·ν Electromagnetic energy can be emitted only in quantized form. Max Planck, 1900 Within the current developmental ambit, the main problems of society today include the issue of how to deal with the huge energy consumption that is largely wasted on a daily basis. It should be fully appreciated that energy-saving and energy eciency constitute two of the principal instruments to achieve economic growth and social welfare. Page 49 Chapter 3: Energy-Saving in Smart Environments 3.1 Introduction State eorts to reduce energy wasting in Spain are listed in the "Estrategia Española de Desarrollo Sostenible" approved in 2007 by minister council, and has been put into practice through:  "Estrategia española de cambio climático y energía limpia. Horizonte 2007- 2012-2020" of the Ministry of Environment [77];  "Plan de Acción 2008-2012 de la Estrategia de Ahorro y Eciencia Energética en España" of the Ministry of Commerce, Industry and Tourism, [75];  the new "Plan de Ahorro y Eciencia Energética 2008-2011" of the Ministry of Commerce, Industry and Tourism [76]; and  the Ministry of Economy and Competitiveness, which has supported sustainable development projects, such as that of the AFRISOL buildings, which strives towards the construction of totally renewable buildings in a cooperative way with companies, universities and research centers. In this context, the research provided in this thesis, presented in the paper included within this document in Part II, adapts the utilization of new technologies existent in other domains, in order to improve consumption. Consequently, daily activities, such as switching on lighting, can be adjusted in order to save energy. A software architecture denition (see Chapter 10) forms the foundation upon cornerstone from which these objectives are accomplished. Buildings in Spain today present a sum total of 3,500 million square metres, and account for 17% of total energy consumption [46]. European authorities approved the new Technical Building Code that requires the installation of solar panels and the use of solar, thermal and photovoltaic energy in new buildings. However, these measures cannot succeed in attaining any real energy saving; they can only satisfy the growing energy demand in each building. This work focuses on how to minimize Page 50 3.2 WSN to sense Smart Environments energy consumption in lighting equipment to obtain a real saving of energy by regulating the resources to attain optimal lighting in buildings. In this sense, the Spanish Association of Home Automation (CEDOM) with the support of the Spanish Institute of Energy Diversication and Saving (IDAE) has published research on energy saving in the residential sector through the installation of home automation [25]. It concluded that energy-saving can be increased to as much as 80% of the present consumption in lighting, 25% in conditioning, and 20% in domestic appliances. The ideas proposed for the reduction in energy consumption and improvement in eciency are aimed at developing partial solutions for the scope of energy consumption in residential buildings. Consequently, an intelligent control of residential buildings is composed of lighting control via lighting sensors. 3.2 WSN to sense Smart Environments Saving energy in smart environment systems is one of the main goals of smart environment research. New approaches to saving energy have arisen thanks to the introduction of Wireless Sensor Networks (WSNs): large networks of embedded devices, containing microcomputers, radios, and sensors. Within the objectives of this study, WSNs are employed to retrieve data on lighting conditions. Several approaches have been proposed in [53, 79] to save energy in this scenario. Nevertheless, the former set of approaches fails to consider user preferences on illuminance, and hence a constant value of illuminance is considered as suitable for every single inhabitant, while the latter approach maintains no knowledge about any type of inhabitants' preferences. Other approaches not based in WSNs, like [86] based on embedded micro-controllers, cannot store any knowledge about inhabitants, and hence it remains unadaptable. Our approach considers inhabitant preferences about lighting and learns them in order to automatically adjust lighting to satisfy these preferences. The aforementioned Technical Building Code establishes 400 lumens as the optimal quantity of light for a standard oce, but makes no mention of how it should be measured. A theoretical computation of how many lumens are provided Page 51 Chapter 3: Energy-Saving in Smart Environments by our articial lighting setup can be made. To this end, a theoretical and empirical analysis about how natural lighting aects indoor lighting is carried out and explained in the journal paper A Study on Saving Energy in Articial Lighting by Making Smart Use of Wireless Sensor Networks and Actuators included in Part II. Moreover, measurements may vary depending on where sensors are located, and upon many other variables, such as type of lights, size of the windows, number of windows, and which direction the windows face. Our approach avoids these diculties, and makes an analysis of the lighting conditions in situ and in real time. The implementation of the experimental case study is based on the ideas presented in [36, 39] who propose a multi-agent approach to control Smart Environments and a paradigm of design based on learning and prediction. The case study stores information according to a model for Smart Environments (see Chapter 9), divided into four categories: device related, inhabitant related, environment related and background. The devices used are the following: a motion sensor to detect whether someone is in the room, an actuator on the uorescent lights to switch them on and o, and a couple of Sentilla Tmotes sensors. Let us describe this last device. Sentilla Tmotes are devices which measure the quantity of light. The lux (symbol: lx) is the SI unit of illuminance and luminous emittance, and can be detected by either of two dierent photodiodes, as explained in Hamamatsu [58]:  S1087 photodiode . For the visible range of the spectrum, from 320 to 730 nanometers. This range is often called Photosynthetically Active Radiation (PAR). This region corresponds with the range of light visible to the human eye.  S1087-01 photodiode . For the visible range up to infrared, from 320 to 1100 nanometers. This is called Total Solar Radiation (TSR). Only the PAR unit is considered in this work. The dimensions of the Sentilla Page 52 3.2 WSN to sense Smart Environments Tmotes are of width 8 cm and height 3.2 cm, and are therefore suciently small to suit ubiquitous applications and non-intrusive systems. These devices include a humidity and temperature sensor, whose information is also retrieved for future improvement and expansions. In Figure 3.1, the Sentilla Tmote module used in our experiment is shown. Low Power Wireless Sensor Module Moteiv Corporation Tmote Sky : Datasheet (11/13/2006) Page 3 of 28 Module Description The Tmote Sky module is a low power “mote” with integrated sensors, radio, antenna, microcontroller, and programming capabilities. USB Connector User Button Reset Button Photosynthetically Active Radiation Sensor (optional) Total Solar Radiation Sensor (optional) 10-pin expansion connector 6-pin expansion connector Internal Antenna CC2420 Radio SMA Antenna Connector (optional) Humidity Temperature Sensor (optional) USB Transmit LED USB Receive LED LEDs USB Microcontroller Digital switch Isolating USB from microcontroller JTAG connector USB Connector User Button Reset Button Photosynthetically Active Radiation Sensor (optional) Total Solar Radiation Sensor (optional) 10-pin expansion connector 6-pin expansion connector Internal Antenna CC2420 Radio SMA Antenna Connector (optional) Humidity Temperature Sensor (optional) USB Transmit LED USB Receive LED LEDs USB Microcontroller Digital switch Isolating USB from microcontroller JTAG connector USB Flash (2kB) ST Code Flash (1MB) Texas Instruments MSP430 F1611 microcontroller 32kHz oscillator 48-bit silicon serial ID 2-pin SVS connector USB Flash (2kB) ST Code Flash (1MB) Texas Instruments MSP430 F1611 microcontroller 32kHz oscillator 48-bit silicon serial ID 2-pin SVS connector Figure 1 : Front and Back of the Tmote Sky module Figure 3.1: Front of the Sentilla Tmote module. The connectivity of the Sentilla Tmote with other motes and computers is executed through the IEEE 802.15.4 (ZigBee) protocol, which minimizes battery consumption. The mesh network protocol implemented by Tmotes software is chosen. In this way, a network to share information and forward it to reach a wider distribution can be designed for the motes. Nowadays Sentilla Tmotes are packaged in a development kit, including an IDE based on Eclipse 3.2 for development in Java. The hardware implements a Java Runtime Environment which can run various applications for the retrieval, processing and transmission of data from sensors. Page 53 PART II Selected research works Page 55 Smart Scheduling for Saving Energy in Grid-Computing Overview This paper addresses the problem of developing Energy-Saving policies for Grid- Computing Environments. Several policies are developed, tested and then compared, thanks to a software tool called Grid'5000 Toolbox, and results are shown. Context This research topic was initiated during the research stages that this PhD candidate carried out at Laboratoire de l'Informatique du Parallélisme of the Ecole Normale Supérieure de Lyon, where permanent researcher Laurent Lefèvre leads the research team on saving energy in Grid-Computing, and on other areas. Thanks to these stages and the help of Laurent Lefèvre and Anne-Cécile Orgerie, these issues could be tackled. This paper is the result of over two years' work in this area. Journal information Figure 4.2: ESWA Cover. The Journal Expert Systems with Applications was selected for the submission of this paper. This Journal is indexed in JCR with an Impact Factor of 2.203 and a 5-Year I.F. 2.455. The Journal stands in ranking Q1 in three categories: Engineering, Electrical & Electronic (41/244); Operations Research & Management Science (5/77); Computer Science, Articial Intelligence (22/111). Page 57 Switch Off Ts policy is the only policy that accesses the agenda for future reservations before deciding what to do. This fact explains why the percentage of prediction success is the greatest. It also returns very good results in terms of energy saving, but with Fig. 9. Statistics tab presenting results for a batch of simulations. Table 1 Selected results for Bordeaux from 1st January to 30th June, sorted according to energy consumed. Energy policy Arranging policy Boot. + Shutt. Energy consumed (kW h) Simulated vs Saved per shutt. Name Params Min Curr Alwz Off n/a SA 1,668,900 149,396 100.13 68.59 0.04 Load [0.9] SA 1,569,934 150,304 100.74 69.01 0.04 S.Off Ts n/a SA 653,856 150,527 100.89 69.11 0.10 Exp. [0.9, 16] SA 1,047,198 151,733 101.70 69.67 0.06 S.Off Rdm n/a SA 890,195 157,645 105.66 72.38 0.07 Gamma [0.9, 1] SA 183,850 186,929 125.29 85.82 0.17 Alwz On n/a SA 0 217,921 146.06 100.05 0.00 Table 2 Selected results for Bordeaux from 1st July to 31st December, sorted according to energy consumed. Energy Policy Arranging policy Boot.+ Shutt. Energy consumed (kW h) Simulated vs Saved per shutt. Name Params Min Curr Alwz Off n/a SA 2,365,598 169,364 100.80 73.61 0.03 S.Off Ts n/a SA 1,584,070 169,894 101.11 73.84 0.04 Load [0.9] SA 2,104,456 170,109 101.24 73.93 0.03 S.Off Rdm n/a SA 1,321,702 173,323 103.15 75.33 0.04 Exp. [0.9, 1] SA 521,878 175,945 104.71 76.47 0.10 Gamma [0.9, 1] SA 957,784 193,180 114.97 83.96 0.04 Alwz On n/a SA 0 230,093 136.94 100.00 0.00 9448 A. Fernández-Montes et al./Expert Systems with Applications 39 (2012) 9443–9450 a much lower number of power cycles: about 40% of the number of power cycles of the Always Off policy. This implies a great advantage over Always Off and Load-0.9 policies. Statistical-based policies, Exponential and Gamma, perform reasonably well in terms of the percentage of prediction success (as expected), particularly the Gamma policy. Energy saving results for the Exponential policy are a bit worse than previous policies, and quite high in terms of power cycles. On the other hand, the Gamma policy performs modestly in terms of energy saving, although the number of power cycles is the lowest, just 11% of Always Off, and hence energy saved per shutting down is the greatest, and the stress imposed on the hardware is the lowest. With respect to the second period, similar results are found. Thus, the minimal energy consumable is 168,024 kW h for a total of 271,149 deployed jobs, and the current energy consumed by Grid’5000 is 230,087 kW h. It is worth noting that in this case, the Exponential policy achieves better results than the Gamma policy. This fact can be explained by taking into account that the number of jobs deployed during the first period is one quarter of the number of jobs deployed during the first period. 6.2. Scheduling arranging policies comparison A set of energy policies has been selected in order to compare results between the two arranging policies: Do Nothing and Simple Aggregation of Jobs. Results are shown in Table 3. Notice that the latter policy, Simple Aggregation of Jobs is consistently the best since it provokes fewer power cycles, saves more energy, and the energy saved per shutting is increased in general. Therefore, if jobs are arranged, even with simple policies, the results are much better than allowing users to choose resources. 6.3. Environmental and Economic results In order to summarize energy-saving results, the costs and CO 2 savings are computed for Bordeaux in Table 4. For the computation of these values, a price of 0.14 euros per kW h and a CO 2 generation of 0.234 kg per kW h are considered. Note that 18,000 euros and 35 tons of CO 2 per year can be saved by implementing the best energy saving policy. Hence, extrapolating results to all 9 locations of Grid’5000, (Bourdeaux, Grenoble, Lille, Lyon, Nancy, Orsay, Rennes, Sophia-Antipolis and Toulouse), 162,000 euros and 318 tons of CO 2 per year could be saved. In terms of energy, up to 129,254 kW h could be saved for Bordeaux and by extrapolating this result to all 9 locations, 1,163,286 kW h. To illustrate how large this quantity of energy really is, it is equivalent to 78 journeys of AVE (high speed rail) train from Madrid to Barcelona, and it is equivalent to the daily energy consumption of 61,314 citizens in the Euro zone. 7. Conclusions and Future work Various methodologies for tackling energy saving in grid computing environments, which could easily be applied to data centres and massive computing environments are presented. We have empirically proven that a suitable policy in grid computing could save a considerable mount of energy and reduce the pollution of CO 2 in the atmosphere. The authors are planning to apply these techniques to these environments in the future, in addition to contributing towards the improvement of energy and arranging policies and their adjustment to new computing environments. A recent upgrade of Grid’5000 toolbox enables us to retrieve data from the cluster located at Centro Informático Científico de Andalucía (CICA), Spain, and hence future work will include external sites for comparison with Grid’5000 sites. Acknowledgement This research is supported by the Spanish Ministry of Science and Innovation R& D project ARTEMISA (TIN2009–14378-C02–01) References Ruth, S. (2009). Green IT more than a three percent solution? IEEE Internet Computing, 13, 74–78. Table 3 Comparison of the two arranging policies comparison for Bordeaux and a selected set of energy policies. Energy policy Arranging policy Boot.+ shutt. Energy consumed (kW h) Simulated vs Saved per shutt. Name Params Min Curr Load [0.9] SA 1,569,934 150,304 100.74 69.01 0.04 Load [0.9] DN 1,561,122 154,141 103.31 70.77 0.04 Load [0.6] SA 902,932 158,205 106.03 72.64 0.07 Load [0.3] SA 299,304 177,839 119.19 81.65 0.13 Load [0.6] DN 892,656 186,406 124.94 85.58 0.04 Load [0.3] DN 292,025 208,141 139.50 95.56 0.03 Load [0.0] SA 650 216,386 145.03 99.35 2.18 Load [0.0] DN 650 216,531 145.13 99.42 1.96 S.Off Rdm n/a SA 890,195 157,645 105.66 72.38 0.07 S.Off Rdm n/a DN 879,681 183,073 122.70 84.05 0.04 Table 4 Summary of environmental and economic results for Bordeaux. Energy policy Energy consumed (kW h) Saved energy (kW h) Euros saved CO 2 kg saved Energy Consumed (kW h) Saved energy (kW h) Euros saved CO 2 kg saved Euros saved CO 2 kg saved Alwz S.Off 149,396 68,525 9593 18,776 169,364 60,729 8502 16,640 18,096 35,416 Load 150,304 67,617 9466 18,527 170,109 59,984 8398 16,436 17,864 34,963 S.Off Ts 150,527 67,394 9435 18,466 169,894 60,198 8428 16,494 17,863 34,960 Exp. 151,733 66,188 9266 18,135 175,945 54,148 7581 14,836 16,847 32,972 S.Off Rnd 157,645 60,276 8439 16,516 173,323 56,770 7948 15,555 16,386 32,070 Gamma 186,929 30,992 4339 8492 193,180 36,912 5168 10,114 9507 18,606 Alwz On 217921 0 0 0 230093 0 0 0 0 0 First period Second period Total A. Fernández-Montes et al./Expert Systems with Applications 39 (2012) 9443–9450 9449 Sanchez, M., Wong, H., Berard, S., & Koomey, J. (2011). Implications of historical trends in the electrical efficiency of computing. IEEE Annals of the History of Computing, 33, 46–54. Koomey, J. G. (2008). Worldwide electricity used in data centers. Environmental Research Letters, 3, 034008. Ren, G., Tune, E., Moseley, T., Shi, Y., & Hundt, R.S.R. (2010) Google-Wide profiling: a continuous profiling infrastructure for data centers, Micro IEEE 30, 65–79. Tseng, L.-Y., Chin, Y.-H., & Wang, S.-C. (2009). A minimized makespan scheduler with multiple factors for Grid computing systems. Expert Systems with Applications, 36, 11118–11130. Orgerie, A.-C., Lefèvre, L., Gelas, J.-P.(2008). Save Watts in your grid: green strategies for energy-aware framework in large scale distributed systems. In: 2008 14th IEEE International Conference on Parallel and Distributed Systems, pp.171–178. Idinfor, http://madeira.lsi.us.es/GrupoIdinfor, 2011. Grid Toolbox, http://madeira.lsi.us.es/proyectosWeb/grid.php, 2011. Martin-Michiellot, S. (2008).JScience, a general scientific API in Java, Practice. Reese, G. (2000). Database Programming with JDBC and Java (2nd ed.). O’Reilly & Associates, Inc.. Keith, M., & Schincariol, M. (2006). Pro EJB 3: Java Persistence API (Pro). Berkely, CA, USA: Apress. Bertie, A. (2002). Java applications for teaching statistics. MSOR Connections, 2, 78–81. Gupta, S. (2011). Pro Apache Log4j (2nd ed.). Apress. Khan, A. (2010). JExcel: A Java library for reading/writing Excel, Accessed Online at http://jexcelapi.sourceforge.net/. Green Grid, T. (2011). Recommendations for measuring and reporting overall data center efficiency Version 2 – Measuring PUE for Data Centers. 9450 A. Fernández-Montes et al./Expert Systems with Applications 39 (2012) 9443–9450 Evaluating decision-making performance in a Grid-computing environment using DEA Overview This paper addresses the problem of comparing eciency between locations of a Grid-Computing infrastructure. Eciency is compared thanks to DEA techniques, and the results are shown. Context This research topic was based on previous knowledge from PhD Francisco Velasco Morente from the department of Applied Economics of the University of Seville. Thanks to his help, these issues could be tackled. This paper is the result of over one year of work in this area and applies usage information (and data from other sources) from simulations performed for a previous paper titled Smart Scheduling for Saving Energy in Grid-Computing . Journal information Figure 5.2: ESWA Cover. The Journal Expert Systems with Applications was selected for the submission of this paper. This Journal is indexed in JCR with an Impact Factor of 2.203 and a 5-Year I.F. 2.455. The Journal stands in ranking Q1 in three categories: Engineering, Electrical & Electronic (41/244); Operations Research & Management Science (5/77); Computer Science, Articial Intelligence (22/111). Page 69 Evaluating decision-making performance in a grid-computing environment using DEA A. Fernández-Montes a, ⇑ , F. Velasco b , J.A. Ortega a a Departamento de Lenguajes y Sistemas Informáticos, Universidad de Sevilla, Spain b Departamento de Economía Aplicada I, Universidad de Sevilla, Spain article info Keywords: DEA Energy policies Efficiency Performance Return-to-scale abstract Energy saving involves two direct benefits: sustainability and cost reduction, both of which Information Technologies must be aware. In this context, clusters, grids and data centres represent the hungriest consumers of energy. Energy-saving policies for these infrastructures must be applied in order to maximize their resources. The aim of this paper is to compare how efficient these policies are in each location of a grid infrastructure. By identifying efficient policies in each location and the slack in inputs and outputs of the inefficient locations, Data Envelopment Analysis presents a very useful technique for comparing and improving efficiency level. This work enables managers to uncover any misuse of resources so that corrective action can be taken. Ó2012 Elsevier Ltd. All rights reserved. 1. Introduction Data Envelopment Analysis (DEA) is a nonparametric method to provide a relative efficiency assessment (called DEA efficient) for a group of decision-making units (DMU) or for productive efficiency (aka technical efficiency) with a multiple number of inputs and outputs. DEA was first proposed in Charnes, Cooper, and Rhodes (1978) and is commonly used in operations research and economics to empirically measure productive efficiency of DMUs. In order to determine whether a DMU is efficient is as easy as checking if the DMU is on the ‘‘frontier’’ of the production possibility set. In this way, DEA identifies a ‘‘frontier’’ on which the relative performance of all utilities in the sample can be compared. In recent years, a great variety of applications of DEA have appeared for the evaluation of the performances of many kinds of entities engaged in various contexts. DEA is especially useful when examining the nature of complex (often unknown) relations between multiple inputs and multiple outputs. DEA has been used both in private (Amirteimoori & Emrouznejad, 2012; Chiang & Hwang, 2010; Eilat, Golany, & Shtub, 2008; Emrouznejad, Parker, & Tavares, 2008) and in public contexts (Afonso, Schuknecht, & Tanzi, 2010; Gonzalez-Rodriguez, Velasco-Morente, & González-Abril, 2010). Regarding energy efficiency studies, DEA is commonly applied for the study and comparison of the performance and efficiency of energy industries, above all in the electricity industry, see (Pérez-Reyes & Tovar, 2009; Pombo & Taborda, 2006; Tovar, Javier Ramos-Real, & de Almeida, 2011; Vaninsky, 2006; Weyman-Jones, 1991). More recently, it has also been applied to IT companies in Serrano-cinca and Fuertes-calle (2005). Recently, it has also been popularized in environmental performance measurement due to its empirical applicability. In this work, DEA is used as a method to compare energy-con- sumption efficiency between each Grid’5000 location, where productive efficiency is measured as the energy consumed to run Grid’5000 jobs at each location. The rest of this paper is structured as follows: Section 2includes a brief introduction to DEA methodology used in this paper. Various on–off policies, designed to save energy are presented, and a comparison between current energy consumption and the results of each on–off policy are given in Section 3. The way in which jobs can be scheduled between resources is shown in Section 4. Software developed for testing and simulation is explained in Section 5and the dataset used for DEA is described and presented. Finally, in Sections 6 and 7, results are given and conclusions are drawn. 2. Data Envelopment Analysis DEA has been successfully applied to several sectors. The method establishes a best-practice production frontier (or envelop) based on the empirical input and output data on DMUs. It determines the level of production inefficiency of a DMU by projecting the unit onto the frontier. The original DEA model, introduced in Charnes et al. (1978), was set up with input orientation and assumes constant returns to scale (CRS). In an input-oriented model, the desired output level is achieved by minimizing the production inputs. The CRS 0957-4174/$ - see front matter Ó2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.eswa.2012.04.028 ⇑ Corresponding author. Tel.: +34 626215333; fax: +34 954557139. E-mail address: [email protected] (A. Fernández-Montes). Expert Systems with Applications 39 (2012) 12061–12070 Contents lists available at SciVerse ScienceDirect Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa assumption suggests that an increase in the amount of inputs utilized would lead to a proportional increase in the amount of outputs generated. The original model has been subsequently extended and numerous variations of DEA. For example, a DEA model can be set up to be output-oriented (Charnes, Cooper, & Rhodes, 1981), which attempts to maximize outputs with a set of available inputs. Another significant development of the DEA model by Banker, Charnes, and Cooper (BCC) (Banker, Charnes, & Cooper, 1984) allows for variable returns to scale (VRS). The VRS assumptionsuggests that an increase in the amount of inputs utilized can lead to a proportional or nonproportional change in the amount of outputs generated (Barkhi & Kao, 2010). 3. Energy policies at a glance Energy policies establish the managing of grid resources. While other research works try to reduce the make-span (Tseng, Chin, & Wang, 2009), the policies shown in this work try to describe and compute what to do with a resource once a job finishes its execution. Thus, each energy policy decides whether to leave a resource switched on or to switch it off depending on the purpose of the policy. The following subsections show energy policies implemented in Grid’5000 Toolbox. 3.1. Always On This is the simplest energy policy. It never switches resources off, under any condition, and hence resources stay idle, waiting for a new job to be run. Grid’5000 is currently running this way, and therefor these consumption results can be used for comparison with other energy policies in order to know how much energy would have been saved. The number of times resources are switched off or on are always zero, and therefor the stress upon the resource is minimal. 3.2. Always Off This policy always switches resources off, under any condition, and hence a resource starts shutting down immediately after any job finishes, and remains switched off. If a new job arrives, resources assigned have to be booted to run that job. This booting is carried out within reservation limits, and hence the user cannot make effective use of the resources until they are booted. This policy is usually the best regarding energy consumption results, but the number of times a resource is booted up and shut down is always maximum, and the stress produced on the hardware components is the highest, which is seldom desirable. 3.3. Switch off randomly This policy randomly switches resources off or leaves them idle by following a Bernoulli distribution whose parameter is equal to 0.5 when a job finishes. Hence, the number of times resources are switched off or left idle tends towards 50%, and results tend to be half-way between those of the Always Off and Always On policies (regarding the times resources are switched off and those of energy consumption). 3.4. Load Load can be defined as the percentage of resources that are On among the clusters of a location. This policy queries this information and leaves resources idle or switches resources off if the load when finishing a job is greater than a certain threshold or less than a threshold respectively. This threshold is a parameter selected from the GUI from 0 to 1. 3.5. Switch off T S T S is defined as the minimum time which ensures an energy saving if a resource is switched off between two jobs (Orgerie, Lefèvre, & Gelas, 2008). T S can be computed as follows: T S ¼E s P Off d tot þE On!Off þE Off!On P Idle P Off where P Off and P Idle refer to the power consumption in watts of a given resource when it is Off and Idle, respectively. E On?Off and E Off?On refers to the required energy in joules for a given resource to boot or switch it off respectively. E S is the energy saved during T S seconds. Finally, d tot =d On?Off +d Off?On , which is the total time a given resource needs for it to be switched off and switched on. This energy policy queries the agenda to check if the next submitted jobs are going to be run in the grid in less than T S . This policy computes the number of resources that are going to be Fig. 1. Configuration tab presenting setup parameters for a batch of simulations. 12062 A. Fernández-Montes et al. / Expert Systems with Applications 39 (2012) 12061–12070 needed within a time period less than T S , and leaves idle or shuts resources down of the job which has just finished, accordingly. In this way, the simulator attempts to minimize the cycles of booting up and shutting down when these cycles are not going to save energy. 3.6. Exponential The Exponential distribution, denoted by Exp(k), describes the time between events in a Poisson process, i.e. a process in which events occur continuously and independently at a constant average rate (1/k). Under the hypothesis that the arrival of new jobs follows an Exponential distribution, this energy policy attempts to predict the arrival of new jobs. Thus, to compute the kparameter, every time a job finishes, then the mean time between the last jobs is computed, denoted by l . Hence, k=1/ l according to the of method of maximum likelihood. The probability of the arrival of a new job is then computed by means of the Exponential cumulative density function (cdf) as cdf ðT s Þ¼1e T s = l . Therefore, given a threshold value: if cdfðT s ÞPthreshold then leave resources Idle if cdfðT s Þ<threshold then switch resources Off  3.7. Gamma The Gamma distribution, denoted by C (h, j ), is frequently used as a probability model for waiting times, and is a more general model than that given by the Exponential. Under the hypothesis that the arrival of new jobs follows a Gamma distribution, this energy policy attempts to predict the arrival of new jobs. The parameters computed every time a job finishes are: Number of resources available, as resourcesAvailable. These are the resources that are Idle and ready to accept new jobs. Mean resources used by the last jobs, as meanResources. The total number of resources used by the last jobs is computed and divided by the number of jobs. The number of last jobs number is a selected window size. Mean duration of these last jobs, as meanDuration. The sum of the duration of the last jobs is computed and divided by the number of the last jobs. The floor of resourcesAvailable/meanResources,asz. The parameters of the Gamma distribution are then estimated as: h=1/meanDuration and j =z+ 1. The probability of the arrival of a new job is then computed by means of the cumulative density function (cdf) with Fig. 2. Statistics tab presenting results for a batch of simulations. A. Fernández-Montes et al. / Expert Systems with Applications 39 (2012) 12061–12070 12063 6.4. Detailed analysis of T s energy policy technical efficiency Toulouse is selected to illustrate this energy policy although it is not the least efficient location for this energy policy. The corrective actions recommended for this location and policy are detailed in Table 7. Toulouse presents a CRS technical efficiency of 0.936 and a VRS technical efficiency of 0.937, and hence in order to achieve overall efficiency and to belong to the efficient frontier it must reduce input but it has no needs of increasing output. This means that the number of bootings and shuttings must be reduced by 55,393 (6%), and most importantly 27 resources (6%) should be removed. The peers for this location are Lyon, Lille and Bordeaux which belong to the segment of the production frontier where Toulouse has to tend. Within these new dimensions, Toulouse will make the most of its resources and will become efficient in the means of production. The other non-efficient locations should be corrected in a similar way. 6.5. Detailed analysis of Exponential energy policy technical efficiency Rennes is selected to illustrate this energy policy although it is not the least efficient location for this energy policy. The corrective actions for this location and policy are detailed in Table 8. Rennes presents a CRS technical efficiency of 0.581 and a VRS technical efficiency of 0.670, and hence in order to achieve overall efficiency and to belong to the efficient frontier it must reduce input and increase the output ‘number of jobs’. This means that the number of bootings and shuttings must be reduced by 1.4 millions (80%), and most importantly 235 resources (32%) should be removed. In addition, these measures have to be followed by an increase of 38,556 (+66%) in the jobs run at this location. The peers for this location are Lyon and Lille which belong to the segment of the production frontier where Rennes has to tend. Within these new dimensions, Rennes will make the most of its resources and will become efficient in the means of production. The other non-efficient locations should be corrected in a similar way. 6.6. Detailed analysis of Gamma energy policy technical efficiency Nancy is selected to illustrate this energy policy although it is not the least efficient location for this energy policy. The corrective actions recommended for this location and policy are detailed in Table 9. Nancy presents a CRS technical efficiency of 0.406 and a VRS technical efficiency of 0.608, and hence in order to achieve overall efficiency and to belong to the efficient frontier it must reduce input and increase output. This means the number of bootings and shuttings must be reduced by 51,909 (33%), and most importantly 190 resources (33%) should be removed. In addition, these measures have to be followed by an increase of 45,857 (+62%) in the jobs run at this location and a reduction of 39,073 kW h (+207%) in energy consumption. The peers for this location are Lyon and Lille which belong to the segment of the production frontier where Nancy has to tend. Within these new dimensions, Nancy will make the most of its resources and will become efficient in the means of production. The other non-efficient locations should be corrected in a similar way. 7. Conclusions The hypothesis that DEA methodology can be useful for the analysis of technical efficiency in Grid computing environments has been proved. Data Envelopment Analysis enables Grid managers to detect which grid locations present the best and worst performance in terms of energy consumption and efficiency. This methodology also enables several energy policies to be analyzed with regard to their behaviour and the potential differences between running a certain policy at one particular location or another. By means of DEA methodology, system managers are armed with knowledge of which locations are underused and hence decisions regarding the switching off of resources and the relocation of underused locations can be made in order to achieve a better utilization of the Grid infrastructure as a whole. Acknowledgements This research is partially supported by the projects of the Spanish Ministry of Economy and Competitiveness ARTEMISA (TIN2009-14378-C02-01) and Simon (TIC-8052) of the Andalusian Regional Ministry of Economy. References Afonso, A., Schuknecht, L., & Tanzi, V. (2010). Public sector efficiency: Evidence for new EU member states and emerging markets. Applied Economics, 42, 2147–2164. Amirteimoori, A., & Emrouznejad, A. (2012). Optimal input/output reduction in production processes. Decision Support Systems, 52, 742–747. Banker, R. D., Charnes, A., & Cooper, W. W. (1984). Some models for estimating technical and scale inefficiencies in data envelopment analysis. Management Science, 30, 1078–1092. Barkhi, R., & Kao, Y.-C. (2010). Evaluating decision making performance in the GDSS environment using data envelopment analysis. Decision Support Systems, 49, 162–174. Bogetoft, P., & Otto, L. (2010). Benchmarking with DEA, SFA, and R (Vol. 157). Springer. Charnes, W., Cooper, W. W., & Rhodes, E. (1978). Measuring the efficiency of decision making units. European Journal of Operational Research, 2, 429–444. Charnes, A., Cooper, W. W., & Rhodes, E. (1981). Evaluating program and managerial efficiency: An application of data envelopment analysis to program follow through. Management Science, 27, 668–697. Chiang, K., & Hwang, S.-N. (2010). Efficiency measurement for network systems IT impact on firm performance. Decision Support Systems, 48, 437–446. Coelli, T. (1996). A guide to DEAP version 2.1: A data envelopment analysis (computer) program by. Frontiers A Journal of Women Studies, 96/08, 1–49. Eilat, H., Golany, B., & Shtub, A. (2008). R&D project evaluation: An integrated DEA and balanced scorecard approach. Omega, 36, 895–912. Emrouznejad, A., Parker, B. R., & Tavares, G. (2008). Evaluation of research in efficiency and productivity: A survey and analysis of the first 30 years of scholarly literature in DEA. Socio-Economic Planning Sciences, 42, 151–157. Gonzalez-Rodriguez, M., Velasco-Morente, F., & González-Abril, L. (2010). La eficiencia del sistema de protección social español en la reducción de la pobreza. Papeles de población, 16, 123–154. Idinfor (2011). <http://madeira.lsi.us.es/GrupoIdinfor>. Orgerie, A.-C., Lefèvre, L., Gelas, J.-P. (2008). Save watts in your grid: Green strategies for energy-aware framework in large scale distributed systems. In 2008 14th IEEE international conference on parallel and distributed systems (pp. 171–178). Pérez-Reyes, R., & Tovar, B. (2009). Measuring efficiency and productivity change (PTF) in the Peruvian electricity distribution companies after reforms. Energy Policy, 37, 2249–2261. Pombo, C., & Taborda, R. (2006). Performance and efficiency in Colombia’s power distribution system: Effects of the 1994 reform. Energy Economics, 28, 339–369. Serrano-cinca, C., & Fuertes-calle, Y. (2005). Measuring DEA efficiency in Internet companies. Decision Support Systems, 38, 557–573. Tovar, B., Javier Ramos-Real, F., & de Almeida, E. F. (2011). Firm size and productivity. Evidence from the electricity distribution industry in Brazil. Energy Policy, 39, 826–833. Tseng, L.-Y., Chin, Y.-H., & Wang, S.-C. (2009). A minimized makespan scheduler with multiple factors for grid computing systems. Expert Systems with Applications, 36, 11118–11130. Vaninsky, A. (2006). Efficiency of electric power generation in the United States: Analysis and forecast based on data envelopment analysis. Energy Economics, 28, 326–338. Weyman-Jones, T. (1991). Productive efficiency in a regulated industry: The area electricity boards of England and Wales. Energy Economics, 116–122. 12070 A. Fernández-Montes et al. / Expert Systems with Applications 39 (2012) 12061–12070 A Study on Saving Energy in Articial Lighting by Making Smart Use of Wireless Sensor Networks and Actuators Overview This paper addresses the problem of developing Energy-Saving policies for Smart Environments. A WSN was deployed in two oces of the University of Seville to sense lighting conditions and to learn about user preferences in order to save energy in lighting appliances. Context This research topic was initiated during the rst years of research by the PhD candidate. Certain topics within Ubiquitous Computing and Ambient Intelligence constituted the main focus of study. This paper is the result of several years of acquiring a wide range of knowledge in these elds. Journal information THE MAGAZINE OF GLOBAL INTERNETWORKING ® November/December 2009, Vol. 23, No. 6 A Publication of the IEEE Communications Society in cooperation with the IEEE Computer Society and the Internet Society ® ® Celebrating 125 Years of Engineering the Future www.comsoc.org Policy-Based Integration of Multiprovider Digital Home Services Figure 6.2: IEEE Network Cover. A special issue of the IEEE Network Magazine in Digital Home Services was selected for the submission of this paper. This Journal is indexed in JCR with an Impact Factor of 2.148 (JCR- 2009), JCR-5 3.529. The Journal stands in ranking Q1 in four categories: Computer Science, Hardware & Architecture (9/49); Computer Science, Information Systems (25/116); Engineering, Electrical & Electronic (37/246), Telecommunications (8/77). Page 83 IEEE Network • November/December 2009 16 0890-8044/09/$25.00 © 2009 IEEE he Earth is at risk of irreversible damage. The greater the increase in world population, the higher the natural resource consumption. There is an abundance of clear evidence of significant changes in climate conditions, which affect ecosystems of flora, fauna, and, of course, humankind. These changes generate an incentive to find ways to better manage natural resources in order to preserve our current quality of life. As in many developed economies, Spanish residential energy consumption continues to rise, accounting for about 20 percent of the nation’s total energy usage. The Spanish authorities approved the new Technical Building Code in accordance with the European Parliament Directive 2002/91/CE, which requires the installation of solar panels and the use of solar, thermal, and photovoltaic energy in new buildings. However, these measures alone will not succeed in attaining real energy saving; they only curb the growing energy consumption in new constructions. The methodology proposed attends to obtain real energy savings by regulating the resources to reach optimal lighting conditions in buildings. Saving energy in smart environment systems is one of the main goals of smart environment research. Wireless sensor networks (WSNs) — large networks of embedded devices, containing microcomputers, radios, and sensors — open new methods and approaches to saving energy. WSNs are used to retrieve data on lighting conditions. Several approaches have been proposed in [1, 2] to save energy in this scenario. Nevertheless, the first approach does not consider user preferences in illuminance; hence, a constant value of illuminance is considered suitable for every single inhabitant, and the last approach maintains no knowledge of inhabitants’ preferences. Other approaches not based on WSNs, like [3] based on an embedded microcontroller, cannot store any knowledge about inhabitants, so it is not adaptable. Our approach considers inhabitant preferences about lighting and learns them in order to automatically adjust lighting to satisfy these preferences. The aforementioned Spanish Technical Building Code establishes 400 lumens as the optimal quantity of light for a standard office, but makes no mention of how it should be measured. A theoretical computation of how many lumens are provided by our artificial lighting setup can be made. To this end, a theoretical and empirical analysis about how natural lighting affects indoor lighting is carried out in the next sections. Moreover, measurements may vary depending on where sensors are located, and on many other variables: type of lights, size of windows, number of windows, which direction the windows face, and so on. Our approach, as put forward later, avoids these difficulties, and makes an analysis of the lighting conditions in situ and in real time. The implementation of the experimental case study is based on the ideas exposed in [4, 5], which propose a multi-agent approach to control smart environments and a paradigm of design based on learning and prediction. The case study stores information according to [6], a model for smart environments divided in four categories: device related, inhabitant related, environment related, and background. We then compute how much consumption can be saved by adjusting lighting to user preferences. This computation is based on the case study of a standard office at the Department of Computer Languages of the University of Seville. Conclusions are drawn in the final section. Theoretical Analysis In order to study how artificial lighting alters lighting conditions in an indoor environment, a mathematical analysis is carried out that relates the luminance of a room with the lights alternatively switched off and on. Let us suppose in this study that the artificial light has constant power (e.g., 100 W). A function y= f(x) is considered where xand ydenote the quantity of light, measured by the same device, with artificial T T Alejandro Fernández-Montes, Luis Gonzalez-Abril, Juan A. Ortega, and Francisco Velasco Morente, University of Seville Abstract This article is focused on adapting lighting conditions to user lighting preferences. A theoretical analysis of lighting conditions is carried out, and a case study is shown by means of the setup of an experimental environment and an empirical analysis of lighting conditions. Finally, a methodology for saving energy, which adjusts luminance to user preferences, is presented, and a study of the consumption results is given. A Study on Saving Energy in Artificial Lighting by Making Smart Use of Wireless Sensor Networks and Actuators FERNANDEZ-MONTES LAYOUT 11/3/09 11:37 AM Page 16 IEEE Network • November/December 2009 17 lights first switched off and then on (both in the same unit). Clearly, the illumination with the lights on is equal to or greater than that with the lights off; therefore, the inequality 0 ≤x≤yholds. Let xmax denote the maximum quantity of light the device can measure. This value is finite, and does not depend on whether lights are on or off; hence, xmax = f(xmax). The function f(x) is an increasing function, so its first derivative is non-negative. The reason for this result is due to theoretical properties of light, and the fact that artificial light has constant power such that for any 0 ≤x1< x2≤xmax, f(0) ≤ f(x1) ≤f(x2) ≤f(xmax). Hence, 0 ≤x≤xmax and 0 ≤ymin ≤y≤ xmax, where ymin = f(0). Note that ymin denotes the minimum quantity of light with the lights on and is reached when, in an indoor environment, only artificial light contributes to the lighting measurements. With respect to the second derivative of f(x), this must be non-positive since if it were positive, the first derivative of f(x) would be an increasing function, which is impossible since artificial light has constant power. On the other hand, let μdenote the lighting threshold of an inhabitant, which means that μis the minimum quantity of light a user considers sufficient to render additional lighting unnecessary. This value depends on the preference of each user and can vary greatly depending on various factors, such as eye color, ocular difficulty, different habits, and different use of the space. Nevertheless, an interval I= [μ– α, μ+ α] is considered in this article due to the uncertainty of the human perception of lighting conditions. Therefore, the most important value is μ+ α, which must be estimated for each user; this value permits us to ensure that the user has sufficient light. The luminance of the artificial light must be such that ymin ≥μ+ αsince it must guarantee that the lighting preferences of the user can always be satisfied. Clearly, the artificial light must attain at least the value μ+ α, since otherwise, the artificial light would have to be changed. This characteristic should be borne in mind when carrying out maintenance of a lighting system. It should be noted that the value ymin depends solely on the power of artificial light. In Fig. 1 an example of function f(x) can be seen that verifies all theoretical conditions given. It is worth noting that the lighting preferences of the user are exceeded in the zones denoted I and II in Fig. 1. Thus, a regulator of light can be installed (zone I), and a motion sensor can then be configured to switch off the light when it is turned on while no people can be detected indoors (zone II). Therefore, electricity consumption can be reduced with these two devices. Nevertheless, some empirical problems must be solved in order to achieve these energy savings. Case Study In this section a particular case is studied that covers all the theoretical aspects mentioned above. Thus, an experimental environment has been designed in order to carry out an empirical study. Experimental Environment A standard office has been set up at the Department of Computer Languages and Systems of Seville University to obtain a real dataset. This office is intended to be used by three people, who could be visited by colleagues or students. The system of lighting consists of natural light from one large window, and four groups of four artificial fluorescent lights (16 in total) of model F18W/154 T8 manufactured by Sylvania [7], which can act as a complement to the natural light or as the sole source of light. In Fig. 2 the distribution and setup of the room can be studied. It can be observed that the geographical position of the city of Seville was taken into account in the design of the building since the window faces south, which maximizes the quantity of light received during the day. This fact alone implies great savings in consumption. It is worth noting that since any room can face any cardinal direction, the consumption saving and lighting threshold may vary, depending on user preferences (e.g., the position of the user’s desk is in front of the windows, and the user prefers to close the blinds and have the light on since the user finds direct sunlight uncomfortable); hence, these factors must be considered when comparing results. For this reason, in order to study the dependence of lighting on these conditions, the need to retrieve data using several devices was placed in the proposed layout. Figure 1. The relationship between lights off and lights on. y=f(x) x Lights off y=x II I Lights on y ymin xmax μ+α Figure 2. Room setup. Mote sensor (outdoor) 4,3m Fluorescent lights (X10) Mote sensor (indoor) Occupance sensor (X10) N FERNANDEZ-MONTES LAYOUT 11/3/09 11:37 AM Page 17 The devices used are the following: a motion sensor to detect whether someone is in the room, an actuator on the fluorescent lights to switch them on and off, and a couple of Sentilla Tmotes sensors. Let us describe this last device. Sentilla Tmotes are devices that measure the quantity of light. The unit of illuminance is a lux and can be detected by either of two different photodiodes, as explained in Hamamatsu [8]: •S1087 photodiode: For the visible range of the spectrum, from 320 to 730 nm. This range is often called photosynthetically active radiation (PAR). This region corresponds with the range of light visible to the human eye. •S1087-01 photodiode: For the visible range up to infrared, from 320 to 1100 nm. This is called total solar radiation (TSR). Only the PAR unit is considered in this article. The dimensions of the Sentilla Tmotes are 8 cm ×3.2 cm, so they are sufficiently small to suit ubiquitous applications and nonintrusive systems. These devices also include a humidity and temperature sensor, whose information is also retrieved for future improvement and expansions. In Fig. 3 the Sentilla Tmote module used in our experiment is shown. The connectivity of the Sentilla Tmote with other motes and computers is executed through the IEEE 802.15.4 (ZigBee) protocol, which minimizes battery consumption. The mesh network protocol implemented by Tmotes software is chosen. In this way, a network to share information and forward it to reach a wider distribution can be designed for the motes. Nowadays Sentilla Tmotes are packaged in a development kit, including an integrated development environment (IDE) based on Eclipse 3.2 for developing in Java. The hardware implements a Java Runtime Environment, which can run different applications for retrieving, processing, and sending data from sensors. As can be seen in Fig. 2, two Sentilla Tmotes are installed, one indoor and another outdoor, although only those measurements given by the indoor Tmote are taken into account in this article. User Preference Threshold Furthermore, the threshold μof an inhabitant needs defining. This threshold is the minimum quantity of light a user considers sufficient to stay in the room; hence, an experiment with four different inhabitants labeled AFM, JAN, IN (who share the office described in Fig. 2), and JAA (who works in a single office) is carried out. Each inhabitant independently completes a questionnaire about lighting conditions in the room every two hours for ten working days. The question is: “Is this quantity of light enough for you?”; each inhabitant answered either Yes or No. A complete threshold analysis can be found in [9]. According to the questionnaire, the preferences for each inhabitant are: •AFM: The lowest satisfying value for PAR is 90 luxes, and the greatest non-satisfying value for PAR is 83 luxes, that is, μ+ α= 90 and μ– α= 83 for this user. •JAN: The lowest satisfying value for PAR is 95 luxes, and the greatest non-satisfying value for PAR is 88 luxes. •IN: The lowest satisfying value for PAR is 89 luxes, and the greatest non-satisfying value for PAR is 75 luxes. •JAA: The lowest satisfying value for PAR is 101 luxes, and the greatest non-satisfying value for PAR is 92 luxes. Empirical Analysis An analysis of the lighting data retrieved in the environment above is carried out. The main objective in this experiment is to study lighting conditions when lights are switched on and when lights are off. The data is obtained from an indoor mote via the S1087 photodiode (i.e., the PAR is used). Tests, which include switching on and off lights at different moments of the day and night and pairing the luminance measures, are carried out 50 times. Let xand ybe the retrieved data (the full dataset is available in [9]) of the luminance when lights are off and on, respectively. In order to achieve a function that relates the xand ydata, the linear correlation coefficient ris calculated. Since r= IEEE Network • November/December 2009 18 Figure 3. Front of the Sentilla Tmote module. Total radiation sensor (optional) 6-pin expansion connector 10-pin expansion connector Internal antenna SMA antenna connector (optional) CC2420 radio USB microcontroller JTAG connector LEDs USB receive LED Digital switch isolating USB from microcontroller Humidity temperature sensor (optional) Photosynthetically active radiation sensor (optional) User button Reset button USB connector USB transmit LED FERNANDEZ-MONTES LAYOUT 11/3/09 11:37 AM Page 18 IEEE Network • November/December 2009 19 0.9827 is near 1, a linear model is obtained by using the least squares methods (Fig. 4): y= f(x) = 1.1039x+ 97.15. It is worth noting that this function is increasing and that ymin = 97.15. The linear regression of xwith respect to yis x= f*(y) = 0.873y– 77.15. That is, if yis denoted PAROn and is known, an estimated value of x, denoted PAROffEstimated, can be obtained. Saving Energy Under the conditions stated above, it can be guaranteed that μ+ αis the quantity of light an inhabitant considers. Therefore, in order to save energy, indoor artificial lights can be adjusted to this μ+ αvalue. The main goal of the current work is to use the knowledge learned to save energy. Data from motes have been retrieved over several months in different periods of the year and data has been stored in a relational database as described in [10]. The methodology proposed used only one part of the dataset. This subset is shown in the list below: •Indoor lighting PAR: This variable represents the quantity of light received from the indoor Sentilla Tmote S1087 photodiode. It is a continuous variable that ranges from 0 lux upward. •Indoor light state: This variable represents the state of the lights of the room received from the X10 appliance module. It is a discrete variable, with value 0 for lights off and 1 for lights on. •Motion: This variable represents the detection of motion sent by the MS13A X10 device. It is a discrete variable, with value 0 for no motion and 1 for motion detected. From this dataset, several statistical analyses of lighting usage are made; for example, how long lights are left on and off, or when the quantity of light (PAR) is greater or less than user preference (μPAR). This analysis is shown in Table 1. As can be observed in the table, the lights are on for 18.84 percent of the total time, that is, about 4.8 hours a day or 33.6 hours a week (non-working days are also included). The most important value to focus on is the number of instances that the lights are on while PAR values are greater than μPAR, since this means that the lighting conditions exceed those necessary to satisfy the preferences of an inhabitant, and hence energy is being wasted. Abnormal behavior of the illuminance sensor is present in the table since 0.15 percent of the instances have a value less than μPAR when the lights are on. These instances should be considered as error values, although the percentage is negligible. These values are computed for the μPAR of the AFM inhabitant (note that TSR analysis has been omitted but can be found in [9]). The next task is to compute the quantity of luxes in all the instances of the database where the lights are on (PAROn): where PAROffEstimated is the estimated PAR value by regression f*(PAROn) with the lights off. The wasted luxes can now be calculated (Fig. 1) as Notice that the maximum between μPAR and PAROffEstimated is subtracted from the illuminance detected by sensors. This is the borderline between zones I and II shown in the theoretical analysis in Fig. 1. The energy consumed by the lights is calculated as follows: totalTimeOn *#tubes *(wattsPerTube/1000) = Total kWh, and the undesired CO2generated is Total kWh *0.274 kg/kWh = total kg of CO2. By supposing a linear relation between generated luxes and their consumption, the total waste of lighting can be computed as about 72 percent of consumption. Making a smart adjustment of the lights could save about 340 kWh, 93 kg of CO2, and €46 (€0.14/kWh) a year per standard office, based on the data retrieved for the case study and the methodology proposed. It is worth noting that the adjustment of lights is not trivial; fluorescent lights are generally suitable for dimming, so a discretization of lights is carried out. A total of 16 fluorescent lights are employed so that some or all of them can be switched on at any time. Moreover, the regression line of the relation PAR On/Par Off can be used (instead of Par Off/Par On as shown in the previous section). In this way, the quantity of luxes (PAROffEstimated) when the lights are off can be estimated before any action is executed. See Table 2 for an analysis of the results. PAR PAR PAR lights On On OffEstimated − () ⎡ ⎣⎤ ⎦ = ∑max , .μ PAR PAR lights On On OffEstimated − () = ∑, Figure 4. PAR Off/PAR On relation. PAR off PAR off/on 100 1000 1200 0 PAR on 800 600 400 200 0 200 300 400 500 600 700 800 900 y=1.1039x+97.15 R2=0.9637 PAR on Lineal (PAR on) Table 1. Study of lighting behavior. Total PAR PAR > μPAR PAR < μPAR #instances %#instances %#instances % Lights on 13,810 18.84 13,710 52.53 70 0.15 Lights off 59,490 81.16 12,390 47.47 47,100 99.85 Total 73,300 100 26,100 100 47,170 100 FERNANDEZ-MONTES LAYOUT 11/3/09 11:37 AM Page 19 IEEE Network • November/December 2009 20 Conclusions In this article it has been shown that electricity consumption can be reduced by incorporating two devices (a regulator of light and a motion sensor) in an indoor setup. Hence, any saving obtained in electricity consumption can be greater if the preferences of users and the environment are set up by following an optimal criterion. Thus, we have proved that in a experimental room, the waste of lighting is about 72 percent of consumption in spite of the optimality of the geographical orientation. Furthermore, the methodology proposed is adaptable to any environment and any user, due to the fact that lighting conditions are not predefined but learned by a smart environment system. Acknowledgments This research is partially supported by the MEC I+D project InCare. Ref: TSI2006-13390-C02-02 and the Andalusian Excellence I+D project CUBICO Ref: TIC2141. References [1] A. D. Galasiu et al ., “Energy Saving Lighting Control Systems for Open-Plan Offices: A Field Study,” LEUKOS , vol. 4, no. 1, July 2007, pp. 7–29. [2] S. Onaygil and O. Guler, “Determination of the Energy Saving by Daylight Responsive Lighting Control Systems with an Example from Istanbul,” Building and Environment , vol. 38, no. 7, 2003, pp. 973–77. [3] J. Parker and R. Wang, “Energy Saving Lighting System for, e.g., Emergency Lighting Backup, has Microcontroller that Controls Power Supply Unit and Light Source Based on Movement of People, Conditions of Battery, and Amount of Light in Predetermined Area,” Patent no. US2009039797-A1, 2009. [4] D. Cook, M. Youngblood, and S. Das, “A Multi-Agent Approach to Controlling a Smart Environment,” in Designing Smart Homes , Lecture Notes in Computer Science , vol. 4008, 2006, p. 165; http://www.springerlink.com/ content/142027w275531607/fulltext.pdf [5] S. Das and D. Cook, “Designing Smart Environments: A Paradigm Based on Learning and Prediction,” in Mobile, Wireless, and Sensor Networks , Wiley- IEEE Press; http://www.springerlink.com/content/7t87564r42285713/fulltext.pdf [6] A. Fernández-Montes et al ., “Modeling Smart Homes for Prediction Algorithms,” in Knowledge-Based Intelligent Information and Engineering Systems , Lecture Notes in Computer Science , vol. 4693, 2007, p. 26; http://www.springerlink.com/content/m6q1051w442nv845/fulltext.pdf [7] Sylvania; http://www.sylvania-lamps.com/catalog/index.php?id=245 [8] Hamamatsu, “Si Photodiode S1087/s1133 Series — Ceramic Package Photodiode with Low Dark Current”; http://jp.hamamatsu.com/products/sensorssd/pd140/pd142/S1087/index_en.html [9] Idinfor; http://idinfor.lsi.us.es [10] A. Fernández-Montes et al ., “Smart Environment Vectorization: An Approach to Learning of User Lighting Preferences,” in Knowledge-Based Intelligent Information and Engineering Systems , Lecture Notes in Computer Science , vol. 5177, 2008, pp. 765–72. Biographies ALEJANDRO FERNANDEZ-MONTES ([email protected]) received his B.Sc. and M.Sc. graduate degrees in computer science from the University of Seville, Spain, in 2006 and 2007, respectively. He is an assistant teacher with the Department of Computing Systems and Languages at the same university, with the Idinfor group. He has worked on several research projects supported by the Spanish government and the European Union. He is a co-author of journal papers, book chapters, and conference papers. His research interests include ubiquitous computing, energy saving, and knowledge discovery. LUIS GONZALEZ-ABRIL ([email protected]) obtained his B.Sc. in mathematics in 1986, and his Ph.D. in economics in 2002 from the University of Seville. He is a senior lecturer in the Department of Applied Economics I at the same university. His research interests are in support vector machines, machine learning, qualitative knowledge, and time series analysis. JUAN ANTONIO ORTEGA RAMIREZ ([email protected]) obtained his Ph.D. in computer science in 2000 from the University of Seville. He has been a senior lecturer in the Department of Languages and Computer Systems at the same university since 1992. His research interests are in time series analysis, machine learning, and global information systems and specifically in the domotic systems. FRANCISCO VELASCO MORENTE ([email protected]) obtained a Ph.D. in mathematics in 1991 from the University of Seville. He is a senior lecturer in the Department of Applied Economics I at the same university. His research interests are in support vector machine, machine learning, dynamic systems, and location analysis. Table 2. Light consumption and wasting per interval and year. Interval Per year Days computed 101.8 365 Days with lights on 19.2 68.77 Total Luxes generated 167,010,000 598,775,280 Total Luxes wasted 119,715,110 429,210,527 kWh CO2(kg) Euros kWh CO2(kg) Euros Total consumption 132.6 36.3 17.9 475.32 130.24 64.24 Total waste 95.0 26.0 12.8 340.72 93.36 46.05 FERNANDEZ-MONTES LAYOUT 11/3/09 11:37 AM Page 20 54 PERVASIVE computing Published by the IEEE CS n 1536-1268/09/$25.00 © 2009 IEEE Pervasive Computing Approaches to Environmental Sustainability n n n n n n n n n n n n n A Decision support system for crop mAnAgement Rolando A. Cardenas-Tamayo and J. Antonio García-Macías, CICESE Research Center The agricultural sector is one of the most important sources of income and production worldwide. Agriculture is directly related to sustainability issues such as water availability and soil conservation, so using these resources efficiently is important. Continuous monitoring systems can maximize their proper use. Such environments are highly dynamic, and systems that support the decision-making process are valuable tools. However, these systems need large amounts of information that must be provided continuously; additionally, they require constant attention by their users. Therefore, we designed and implemented a decision-support system for monitoring crops using pervasive computing technologies such as wireless sensor and actuator networks. The prototype we implemented includes tools that provide real-time information about crop status and the surrounding environment, contributing to the use of techniques such as fertigation (the application of fertilizers, soil amendments, or other water-soluble products through an irrigation system). We carried out an experimental evaluation of our prototype with a group of potential users based on the technology acceptance model (TAM), gathering the users’ perceptions about usefulness, ease of use, and intention. We believe that our proposal has the potential to reduce costs and, by using precise information, improve resource management for crop production. We’ve begun long-duration field tests that will enable a better assessment of our proposal. For more information, contact the authors at [email protected] or [email protected]. n n n n n n n n n n n n n ADApt: AuDience Design of persuAsive technology Timothy M. Miller, Patrick Rich, and Janet Davis, Grinnell College Although many recent research efforts have used pervasive computing to persuade people to use less energy in the course of their everyday activities, stakeholders have rarely been involved in those design efforts. In response, we held a series of participatory design events at Grinnell College to involve potential users in designing an ambient persuasive display that targets an energy-consuming behavior. We used an exploratory game to help participants find places where they consume. We then introduced them to sensor and actuator technology through the Phidgets rapid prototyping framework (www.phidgets.com), engaged them in building mockups, and invited them to critique prototypes of several designs. Ultimately, we designed an interactive sculpture as part of a staircase in the science building, intended to attract people to the stairs from a nearby elevator. The system uses glowing LEDs embedded in hand-sized wire sculptures to evoke fireflies along the walls and stair railings. Participants can “catch” the fireflies, which are equipped with vibration sensors, by tapping them—thus triggering a celebratory animation. Users can race EdItor’s Intro In this issue’s Works in Progress department, we have eight projects with a focus on environmental sustainability. the first three projects explore sensing and pervasive computing techniques for monitoring environmental conditions in outdoor situations. the next four projects use pervasive computing in indoor environments to inform individuals about their energy and resource consumption with the goal of positively influencing their behaviors. the final project aims to develop an energygeneration infrastructure that combines multiple types of renewable energy sources. —Anthony D. Joseph Works in Progress Editor: Anthony d. Joseph n UC Berkeley n [email protected].edu JAnUAry–MArCh 2009 PERVASIVE computing 55 the fireflies up the stairs, showing that most people can walk faster than the nearby elevator. A smaller display near the elevator directs people to the sculpture in the stairway. We’re currently gathering baseline data on stair and elevator use and preparing to install the system. We plan to evaluate the system with respect to frequency of stair and elevator use, frequency of interactions with the system itself, and self-reports regarding the system’s effectiveness in promoting behavior and attitude change. For more information, contact Janet Davis at [email protected]. n n n n n n n n n n n n n reAlnet: An environmentAl Wireless sensor netWork Joan Albesa, Manel Gasulla, Jorge Higuera, María Teresa Penella, and José Garcia, Universitat Politècnica de Catalunya Wireless sensor networks (WSNs) have recently emerged as a feasible technology in several fields, including environmental monitoring and control. We’re deploying an environmental WSN, called REALnet, at our campus to monitor physical parameters from the air, water, and soil and to act on the environment in a more sustainable way. We’re investigating several topics that can be critical barriers to developing WSNs: Power consumption: Implementing new low-power measuring methods; synchronizing nodes by compensating for the clock drifts to reduce duty cycle and power consumption. Power supply of sensor nodes: Energy harvesting as an alternative to primary batteries; using efficient energy and power management techniques. Interoperability: Implementing suitable wireless standards such as IEEE 802.15.4, Zigbee, and 6LoWPAN (IPv6-based low-power wireless personal area network); implement- • • • ing IEEE 1451 on top of them to interoperate the diverse networks. We’ve already deployed five nodes: one coordinator node, powered from the mains; one sensor node, powered by primary batteries, that monitors the campus pond’s water level and temperature; and three router nodes, powered by solar cells, that route data from the sensor node to the central node. Router nodes also transmit data about the level of solar irradiation, temperature, and battery level. We’ll soon add nodes to monitor a tiny weather station and low-power sensors to measure the soil temperature and moisture. We’re implementing the IEEE 1451 standard and will save data on a database server for access through the Internet. For more information, contact Manel Gasulla at [email protected]. n n n n n n n n n n n n n AutomAting smArtenvironment systems Alejandro Fernández-Montes, Universidad de Sevilla This work focuses on learning user preferences about lighting and air conditioning to plan actions that an automated smart-environment system should perform to satisfy these preferences. However, the system must also detect inhabitants’ wasteful behaviors such as turning on lights when natural light is available or abusing the air conditioning system when it’s not necessary. The system must then teach the inhabitants to save energy by avoiding those behaviors. My research group proposes a general software architecture that can guide developers in building smartenvironment solutions with sustainability features. I have empirically tested this architecture, using it to develop a smart-environment solution for learning users’ lighting preferences. This proposal is based on the goals of ubiquitous computing as Mark Weiser proposed them in the late 1980s. Taking this as a starting point, we believe that automation in smart environments should be organized as continuous interaction between three main tasks: perception, reasoning, and acting. The system perceives the environment’s state by means of the physical components distributed throughout the environment. It then reasons about the state and produces a list of possible actions to take. Finally, it carries out these actions, each of which is made up of several processes. A policy manager can guide developers to deploy green policies in order to save energy. To test this general software architecture, we’re developing a smart solution that covers tasks of perception, reasoning, and acting. This is a Java solution built on top of the Sentilla Development Kit. It has been very useful in polishing this architecture proposal. For more information, contact the author at [email protected]. n n n n n n n n n n n n n Wireless sensor netWorks for emergency ApplicAtions Maria-Angeles Grado-Caffaro, Sapienza Studies The European Union’s Sixth Framework Programme is funding the WINSOC project (wireless sensor networks with self-organization capabilities for critical and emergency applications). This research started in September 2006 and is due to finish in February 2009. The project aims to radically improve sensor networks’ performance using a biologically inspired design. The system can self-assemble into organized structures in which the sensor nodes behave as biological entities that are able to make decisions on the basis of their local environments and their own individual states. These nodes are small calculating machines (much simpler than the current ones) in the network that need to carry out only very simple rules, thus eliminating the need for the inefficient, complex 56 PERVASIVE computing www.computer.org/pervasive Works in Progress Works in Progress protocol interactions used for end-to- end communications. This approach lets us build distributed detection and estimation capabilities that are crucial for understanding a WSN scenario and eliminates the need for sending all the data to a fusion center. This technology is attractive for a wide range of environmental purposes; so far, the WINSOC team has analyzed landslide and wildfire applications. In India, our Amrita University partner has deployed sensors to monitor humidity and porosity in the terrain as well as the forces involved in terrain displacements. The system sends this information to a satellite, which then conveys it to a control center. Also, our Czech Center for Science and Society partner has introduced sensors in a forest to detect and locate heat and smoke sources. A simulation of a spreading fire, developed by the WINSOC team, provides monitoring and alerts. Selex Communications (Italy) coordinates the WINSOC project. Other partners include the University of Rome “La Sapienza” (Italy), the Ecole Polytechnique Fédérale de Lausanne (Switzerland), Intracom Telecom (Greece), the Commissariat per l’Energie Atomique-LETI (France), the Czech Center for Science and Society (Czech Republic), Dune (Italy), Universitat Politècnica de Catalunya (Spain), the Indian Space Organization (India), Amrita University (India), and the science and technology consultancy firm Sapienza Studies (Spain). For more information, contact the author at [email protected] or visit the project Web site at www.winsoc.org. n n n n n n n n n n n n n An Ambient WAter consumption DisplAy Karin Kappel and Thomas Grechenig, Vienna University of Technology Water is becoming a scarce resource worldwide. Nevertheless, according to the United Nations, water usage has grown at twice the rate of population growth during the past century. There is an urgent need for people to change their water usage behavior. Ambient displays can help people by giving immediate and direct feedback regarding their consumption while at the same time blending in with the environment. Water consumption for one shower varies heavily depending on the user’s shower habits, so the potential for water conservation is great. We built a prototype of an ambient water consumption display system for the shower, which gives the user feedback and an impression of the amount of water going down the drain (see Figure 1). This impression is achieved by using the metaphor of the drain being closed and the water level increasing within the shower. The imaginary water level is visualized in the form of LEDs that are vertically assembled on a stick. One additional LED lights up for every five liters, thus delivering direct feedback. Two initial tests in two-person households show promising results. A couple used to argue over who used more water. After we installed the display, they learned that the woman used only half as much water even though she spent more time in the shower. This discovery stimulated the man to use even less water. Overall, each subject, depending on the individual goal, made different efforts to act more sustainably after having installed the ambient water consumption display. For more information, contact the authors at [email protected]. ac.at. n n n n n n n n n n n n n hybriD energy proDuction AnD Delivery using multiple reneWAble sources lhan Umut, Kırklareli University Erdem Uçar, Trakya University This study investigates energy production from renewable energy sources such as solar, wind, and water and storage of the produced energy in batteries before consumption. The processes take place simultaneously. The hybrid system shown in Figure 2 will be able to supply energy uninterrupted. When one or more of the resources ceases functioning, the other sources will Figure 1. The shower prototype. An ambient display can help people by giving immediate and direct feedback regarding water consumption. Hybrid system Battery cluster and inverter Water mill Wind turbine Solar tracking Renewable energy sources Figure 2. Schematic view of the hybrid energy production and delivery system. JAnUAry–MArCh 2009 PERVASIVE computing 57 Works in Progress be able to feed the system. The main advantage of this type of system is the energy production’s independence from the electricity network. Therefore, we will be able to supply energy for environmental lighting and security systems, even in rural areas. The system’s solar module is already in use (see Figure 3). We designed this module in such a way that the system follows the sun automatically. The studies for the other modules are continuing. We’ll conduct tests for the other modules once we can find a convenient location where we can produce energy from all three sources. It will also be able to instantly measure the quantity of produced energy and send that data to a computer through a cable or wireless connection. The software will be able to express the amount of produced energy from each source as daily, monthly, and yearly totals. The system will record the data, monitor system status, and control the system. For more information contact the authors at [email protected]. n n n n n n n n n n n n n optimAl DistributeD control for sustAinAble builDings Josh Wall and John Ward, CSIRO With the building sector contributing 30 to 40 percent of global CO2 emissions, sustainable building technologies have been identified as one of the most cost-effective approaches for reducing carbon emissions. Australia’s Commonwealth Scientific and Industrial Research Organization (CSIRO) believes that advances in optimal, distributed building controls will help realize significant savings while maintaining or even improving indoor climates. CSIRO focuses on four key areas in intelligent building controls: optimizing the trade-off between user comfort, operating cost, and carbon emissions; forecasting building conditions; applying localized control responses; and identifying anomalous conditions that might represent faults or inefficiencies. The application of pervasive computing technologies is enabling these developments. Specifically, by using highspatial-resolution mesh networks, we can obtain detailed condition, operation, and energy mappings for use both at a local sensing and control level and when aggregated for whole-building optimization. CSIRO is extending its optimal building control technology to wireless sensor networks to facilitate lowcost, ubiquitous controls. With low energy consumption and robust ad hoc communications, such controls are ideal for retrofits to existing building stock—which is crucial for maximizing carbon reductions. Using distributed, embedded hardware, intelligence can be immersed into building sensor and control networks using localized data as input to optimized adaptive control routines with fast response and low network and computational overhead. This input can be combined with realtime comfort feedback from occupants via pervasive PC-based virtual comfort sensor software (see Figure 4). This technology has the potential to break through traditional comfort barriers by improving thermal satisfaction—not only from the direct physical effect of occupant adjustments on indoor climate but also from the empowerment of the occupants. Supported by ever-evolving software architectures and embedded hardware platforms, possibilities for optimal, distributed control of sustainable buildings are growing dramatically, with significant potential energy savings and reduction of carbon emissions becoming reality. For more information, contact the authors at [email protected]. Figure 3. The solar tracking device and other system components. Wireless sensor/controller Virtual comfort sensor Figure 4. Integration of wireless sensors/controllers and virtual comfort sensors for optimal, distributed control of sustainable buildings. Ignac Lovrek Robert J. Howlett Lakhmi C. Jain (Eds.) Knowledge-Based Intelligent Information and Engineering Systems 12th International Conference, KES 2008 Zagreb, Croatia, September 3-5, 2008 Proceedings, Part I 13 Series Editors Randy Goebel, University of Alberta, Edmonton, Canada Jörg Siekmann, University of Saarland, Saarbrücken, Germany Wolfgang Wahlster, DFKI and University of Saarland, Saarbrücken, Germany Volume Editors Ignac Lovrek University of Zagreb, Faculty of Electrical Engineering and Computing Unska 3, 10000 Zagreb, Croatia E-mail: [email protected] Robert J. Howlett KES International 2nd Floor, 145-157 St. John Street, London EC1V 4PY, UK E-mail: rjho[email protected]g Lakhmi C. Jain University of South Australia, School of Electrical and Information Engineering Mawson Lakes Campus, Adelaide, SA 5095, Australia E-mail: [email protected] Library of Congress Control Number: 2008933454 CR Subject Classification (1998): I.2, H.4, H.3, J.1, H.5, K.6, K.4 LNCS Sublibrary: SL 7 – Artificial Intelligence ISSN 0302-9743 ISBN-10 3-540-85562-9 Springer Berlin Heidelberg New York ISBN-13 978-3-540-85562-0 Springer Berlin Heidelberg New York This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting, reproduction on microfilms or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable to prosecution under the German Copyright Law. Springer is a part of Springer Science+Business Media springer.com © Springer-Verlag Berlin Heidelberg 2008 Printed in Germany Typesetting: Camera-ready by author, data conversion by Scientific Publishing Services, Chennai, India Printed on acid-free paper SPIN: 12511303 06/3180 543210 XXXIV Table of Contents – Part I Construction and Evaluation of a Robot Dance System ............... 682 Kuniya Shinozaki, Akitsugu Iwatani, and Ryohei Nakatsu Development and Evaluation of a Centaur Robot .................... 692 Satoshi Tsuda, Kuniya Shinozaki, and Ryohei Nakatsu Collaboration Options for Small and Medium Size Enterprises ......... 701 Sylvia Encheva and Sharil Tumin Service Robot System Using Personal Attribute and Acquisition Preference Attribute with Networked Robots ........................ 709 Kota Nakamura, Yoshiharu Yoshida, Toru Yamaguchi, and Eri-Shimokawara-Sato Self-Protecting Session Initiation Protocol Stack ..................... 717 Zoran Rusinovic and Nikola Bogunovic Early Forest Fire Detection with Sensor Networks: Sliding Window Skylines Approach ............................................... 725 Kreˇsimir Pripuˇzi´c, Hrvoje Belani, and Marin Vukovi´c Analysing Flight Data Using Clustering Methods .................... 733 Christopher Jesse, Honghai Liu, Edward Smart, and David Brown Visual-Based View-Invariant Human Motion Analysis: A Review ....... 741 Xiaofei Ji, Honghai Liu, Yibo Li, and David Brown Generalized Extreme Value for Smooth Component Analysis in Prediction Improvement .......................................... 749 Ryszard Szupiluk, Piotr Wojewnik, and Tomasz Z  abkowski Context Cookies ................................................. 757 Ram´on Herv´as, Gabriel Chavira, Salvador W. Nava, Vladimir Villarreal, and Jos´eBravo Smart Environment Vectorization: An Approach to Learning of User Lighting Preferences .............................................. 765 Alejandro Fern´andez-Montes, Juan A. Ortega, Luis Gonz´alez, Juan A. ´ Alvarez, and Manuel D. Cruz Author Index .................................................. 773 Smart Environment Vectorization An Approach to Learning of User Lighting Preferences Alejandro Fern´andez-Montes1,JuanA.Ortega 1, Luis Gonz´alez2,JuanA.´ Alvarez1, and Manuel D. Cruz1 1Department of Computer Science, University of Sevilla, Sevilla, Spain {afernandez,ortega}@lsi.us.es, {jaalvarez,mancrudia}@us.es 2Department of Applied Economics, University of Sevilla, Sevilla, Spain [email protected] Abstract. The automation of smart environment systems is one of the main goals of smart home researching. This paper focus on learning user lighting preference, considering a working field like a standard office. A review of the smart environment and devices setup is done, showing a real configuration for test purposes. Suitable learning machine techniques are exposed in order to learn these preferences, and suggest the actions the smart environment should execute to satisfy the user preferences. Learning machine techniques proposed are fed with a database, so a proposal for the vectorization of data is described and analyzed. 1 Introduction Smart home technologies are often included as a part of ubiquitous computing. Mark Weiser [1] outlined some principles to describe Ubiquitous Computing (ubicomp) from which we emphasize that the purpose of a computer is to help you do something else. Home technologies have tried to help home inhabitants since its creation. Nowadays, due to the popularization of computational devices, ubiquitous computing is called to be the revolution to develop smart systems with artificial intelligence techniques. Domoweb [2] was a research project originally developed as a residential gateway implementation over the OSGi (Open Services Gateway Initiative) service platform. Nowadays Domoweb conform a great platform where researchers from different disciplines converges and where we can deploy, develop and test smart home related solutions, due to the component based model, and the service oriented architecture that Domoweb and OSGi supports. This article focuses on modeling smart spaces to apply machine learning techniques. We have focused in the learning of user preferences for the lighting of a space. In order to interact with the space and retrieve these preferences, an office at the department of Computer Languages of the University of Seville has been provided of several devices to accomplish these tasks. Artificial intelligent methods can be supported by this model like machine learning algorithms where is centered this article. Some techniques are presented I. Lovrek, R.J. Howlett, and L.C. Jain (Eds.): KES 2008, Part I, LNAI 5177, pp. 765–772, 2008. c Springer-Verlag Berlin Heidelberg 2008 772 A. Fern´andez-Montes et al. 4. Fern´andez-Montes, A., ´ Alvarez, J.A., Ortega, J., Cruz, M., Gonz´alez, L., Velasco, F.: Modeling Smart Homes for Prediction Algorithms. In: Apolloni, B., Howlett, R.J., Jain, L. (eds.) KES 2007, Part II. LNCS (LNAI), vol. 4693, p. 26. Springer, Heidelberg (2007) 5. Cook, D., Youngblood, M., Das, S.: A multi-agent approach to controlling a smart environment. In: Augusto, J.C., Nugent, C.D. (eds.) Designing Smart Homes. LNCS (LNAI), vol. 4008, p. 165. Springer, Heidelberg (2006) 6. Li, J., Bu, Y., Chen, S., Tao, X., Lu, J.: FollowMe: On Research of Pluggable Infrastructure for Context-Awareness. In: Proceedings of the 20th International Conference on Advanced Information Networking and Applications (AINA 2006), vol. 01, pp. 199–204 (2006) 7. Das, S., Cook, D.: Designing Smart Environments: A Paradigm Based on Learning and Prediction. Mobile, Wireless, And Sensor Networks 8. Leake, D., Maguitman, A., Reichherzer, T.: Cases, Context, and Comfort: Opportunities for Case-Based Reasoning in Smart Homes. In: Augusto, J.C., Nugent, C.D. (eds.) Designing Smart Homes. LNCS (LNAI), vol. 4008, pp. 109–131. Springer, Heidelberg (2006) 9. Hagras, H., Callaghan, V., Colley, M., Clarke, G., Pounds-Cornish, A., Duman, H.: Creating an Ambient-Intelligence Environment Using Embedded Agents (2004) 10. Roy, N., Roy, A., Das, S.: Context-Aware Resource Management in Multi- Inhabitant Smart Homes: A Nash H-Learning based Approach. In: Proc. of 4th IEEE Int’l Conf. on Pervasive Computing and Communications (PerCom 2006) (2006) 11. Choi, J., Shin, D., Shin, D.: Research on Design and Implementation of the Artificial Intelligence Agent for Smart Home Based on Support Vector Machine. LNCS. Springer, Heidelberg 12. Yamazaki, T.: Beyond the Smart Home. In: Proceedings of the 2006 International Conference on Hybrid Information Technology, vol. 02, pp. 350–355 (2006) 13. Jiang, L., Liu, D., Yang, B.: Smart home research. In: Proceedings of 2004 International Conference on Machine Learning and Cybernetics, vol. 2 (2004) Bruno Apolloni Robert J. Howlett Lakhmi Jain (Eds.) Knowledge-Based Intelligent Information and Engineering Systems: KES 2007 - WIRN 2007 11th International Conference, KES 2007 XVII Italian Workshop on Neural Networks Vietri sul Mare, Italy, September 12-14, 2007 Proceedings, Part II 13 Series Editors Jaime G. Carbonell, Carnegie Mellon University, Pittsburgh, PA, USA Jörg Siekmann, University of Saarland, Saarbrücken, Germany Volume Editors Bruno Apolloni Dipartimento di Scienze dell’Informazione Università degli Studi di Milano 20135 Milano, Italy E-mail: [email protected] Robert J. Howlett University of Brighton Centre for SMART Systems, School of Engineering Brighton, BN2 4GJ, UK E-mail: R.J.Ho[email protected] Lakhmi Jain University of South Australia Knowledge-Based Intelligent Engineering Systems Centre SA 5095, Australia E-mail: [email protected] Library of Congress Control Number: 2007934283 CR Subject Classification (1998): I.2, H.4, H.3, J.1, H.5, K.6, K.4 LNCS Sublibrary: SL 7 – Artificial Intelligence ISSN 0302-9743 ISBN-10 3-540-74826-1 Springer Berlin Heidelberg New York ISBN-13 978-3-540-74826-7 Springer Berlin Heidelberg New York This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting, reproduction on microfilms or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable to prosecution under the German Copyright Law. Springer is a part of Springer Science+Business Media springer.com © Springer-Verlag Berlin Heidelberg 2007 Printed in Germany Typesetting: Camera-ready by author, data conversion by Scientific Publishing Services, Chennai, India Printed on acid-free paper SPIN: 12120574 06/3180 543210 Table of Contents – Part II I Intelligence Everywhere Ambient Intelligence An Efficient Indexing Technique for Location Prediction of Moving Objects ......................................................... 1 Dong-Oh Kim, Kang-Jun Lee, Dong-Suk Hong, and Ki-Joon Han Combining Smart Tags and Body Fixed Sensors for Disabled People Assistance ...................................................... 10 Juan Antonio ´ Alvarez, Carlos P´erez, Cecilio Angulo, and Juan Antonio Ortega Geometrical Representation of Quantity Space and Its Application to Robot Motion Description......................................... 18 Honghai Liu, David J. Brown, and George M. Coghill Modeling Smart Homes for Prediction Algorithms .................... 26 A. Fern´andez-Montes, J.A. ´ Alvarez, J.A. Ortega, M.D. Cruz, L. Gonz´alez, and F. Velasco Qualitative Correspondence for Object Tracking Using Dynamic Panorama ....................................................... 34 Farshad Fahimi, Honghai Liu, and David J. Brown Artificial Intelligence Applications in Digital Content A Hybrid System: Neural Network with Data Mining in an e-Learning Environment .................................................... 42 David Wen-Shung Tai, Hui-Ju Wu, and Pi-Hsiang Li A Personalization Recommendation Framework of IT Certification e-Learning System ............................................... 50 Heng-Li Yang, Chen-Shu Wang, and Mu-Yen Chen An Architecture for e-Learning System with Computational Intelligence ...................................................... 58 Marc El Alami, Nicolas Casel, and Denis Zampunieris B. Apolloni et al. (Eds.): KES 2007/ WIRN 2007, Part II, LNAI 4693, pp. 26 – 33, 2007. © Springer-Verlag Berlin Heidelberg 2007 Modeling Smart Homes for Prediction Algorithms A. Fernández-Montes, J.A. Álvarez, J.A. Ortega, M.D. Cruz, L. González, and F. Velasco Departamento de lenguajes y sistemas informáticos. ETSI Informática Avda. Reina Mercedes s/n, Sevilla [email protected], [email protected], [email protected], [email protected], [email protected], [email protected] Abstract. This paper reviews the goals of the Domoweb project and the solutions adopted to achieve them. As a result we enjoy a great support to develop smart home techniques and solutions. As a consequence of the acquired experiences a Smart home model is proposed as a division of four main categories. In relation with the smart home model, we show the essential features a smart environment prediction algorithm should satisfy and a procedure to select relevant information from the model to achieve artificial intelligence based solutions. 1 Introduction Smart home technologies are often included as a part of ubiquitous computing. Mark Weiser [1] outlined some principles to describe Ubiquitous Computing (ubicomp) from which we emphasize that the purpose of a computer is to help you do something else. Home technologies have tried to help home inhabitants since its creation. Nowadays, due to the popularization of computational devices, ubiquitous computing is called to be the revolution to develop smart systems with artificial intelligence techniques. Domoweb [2] is a research project which was originally developed as a residential gateway implementation over the OSGi (Open Services Gateway Initiative) service platform. Domoweb implements the standard services any residential gateway must have like http server, web interfaces, device manager, user manager and other basic services. Nowadays Domoweb conform a great platform where researchers from different disciplines converges and where we can deploy, develop and test smart home related solutions, due to the component based model, and the service oriented architecture that Domoweb and OSGi supports. This article focuses on modeling smart homes and the features the prediction algorithms should implement. A good smart environment model must represent the Modeling Smart Homes for Prediction Algorithms 27 properties, states, attributes and any other characteristic that could be useful in building smart environments solutions as is proposed in section 2. Artificial intelligent methods can be supported by this model like prediction algorithms where is centered this article. The features these algorithms must be aware of are defined in section 3 and finally a procedure to discriminate significant information is released in section 4. 2 Smart Home Model Artificial intelligence algorithms need a solid base of knowledge to work. This fact demands us a great effort for building a model of the smart home and its environment. Other projects have helped us to compose the model [3-5] which has been arranged in four main categories explained in next sections and expounded in table 1. 2.1 Device Related This category is the most obvious one so is related with the main elements in a smart home environment. Ambient intelligence algorithms should be aware of next main fields:  Status. Algorithms must know the current states of devices installed on the smart home. Obviously this is essential for these algorithms, and one of the most important domains to build future predictions.  Location. Devices usually occupy a location for a long time and this location may be useful for ambient intelligence algorithms. The model must be able to consider non-still devices as well, like motorized cleaner robots and others. 2.2 Inhabitants Related Smart home algorithms must be aware of the inhabitants’ status to offer appropriate predictions for any user or for the whole group of inhabitants. On this line we discuss some of the necessary fields to infer inhabitants-aware predictions:  Personal data. This field includes all the data concerning to a particular person like name, age, sex and so on.  Location. Inhabitants can move over the home rooms, so smart home systems have to know where each inhabitant is, and should be able to identify them.  Physical state. This field is related with the illness or injuries that an inhabitant can suffer during his life. Smart home technologies must adapt to this situations and offer appropriate replies.  Mental state. The state of mind of a person can be defined as a temporary psychological state. A depressed inhabitant behavior usually differs from a euphoric one, so smart home must be consistent with these circumstances. 28 A. Fernández-Montes et al. Table 1. Smart Home Model CATEGORIES FIELDS DESCRIPTION EXAMPLE Status Current state the devices are. Sensor temperature is at 25ºC DEVICE RELATED Location Where the devices are. Cleaner robot is in Living room. Personal data Name, Age, sex. Diane is 45. Location Where the inhabitants are. Mark is at bedroom. Physical state Illness, injuries, and others states inhabitants are. Roy has a cold. INHABITANTS RELATED Mental state Psychological state the inhabitants are. David is depressed. Date, time Temporary information Current time is 13:36. ENVIRONMENT RELATED Environmental conditions the phenomena that currently occur in the atmosphere It’s rainy Inert entities location Where these entities are. Sofa is at living room. HOME BACKGROUND Home limits properties The properties of the home structure and limits. Bedroom window opacity is 70% 2.3 Environment Related This category probably is the most diffused due to it covers heterogeneous and difficult-to-limit fields as we discuss in the following list: Modeling Smart Homes for Prediction Algorithms 29  Date, time, season. Obviously smart home behavior should be different in different temporal conditions and it may depend in these temporal factors as the air condition policies will differ between summer and winter.  Environmental conditions. This field comprises current environmental conditions (sunny, cloudy, rainy and others). Smart home should make a request for a weather forecast as well, which could be significant to assess future decisions. 2.4 Home Background This category must comprise all the relevant things concerned to inert entities and its properties or qualities. This could be the less relevant category discussed, but anyway could be significant in some concrete applications. We propose a couple of fields related in next listing:  Furniture location and position. Furniture occupies space at home and can be moved. Location (room where the furniture is) and position (place where the furniture is placed in a room) should be known by the smart home systems due to it could be useful by concrete applications like robot movement related algorithms, or presence detection related algorithms.  Home limits properties. The texture of a floor, the color of a wall, or the opacity of the windows could be significant in several cases such as temperature adjustment applications. 3 Features of Prediction Algorithms In this section we present the features that a smart home system must implement, specially related with the prediction algorithms the systems may have. The article doesn’t focuses on artificial intelligence techniques like the studies of [5-10] do, but on what are the most important and indispensable features that must be considered to develop prediction algorithms. Much of the ideas presented below could be useful to implement others smart home algorithms. 3.1 Prediction Supported by Last Events and States Prediction algorithms should consider as input data two main aspects. First it should analyze the last events occurred in the home’s performance field which have changed the home status in any manner. Second it should analyze current state and previous states as well. This way the algorithms should determine last changes in the home status, and which events have been involved in these changes. In section 5 we discuss further about these aspects and propose a way to consider states and events. 30 A. Fernández-Montes et al. 3.2 Predictable by the Inhabitants Smart homes should learn inhabitant behaviors and habits, and build predictions, but it has to be predictable by the inhabitants to achieve no unexpected behaviors. 3.3 Understandable Decisions Prediction algorithms have to offer an explanation of their predictions and/or actions. This way the inhabitants will be more trusted with the decisions the system took. This will improve the user acceptance of the smart home predictions. 3.4 Wrong Decisions Detection and Related Improvement If the smart home executes a wrong decision it should be aware of this failure and be able to learn from its errors. A possible scenario should be when someone arrives home at night and the system orders the hall lights to switch on, but immediately the user performs the opposite action (which should be switch off the lights). The system has to notice this failure and extract some knowledge from this experience to face future similar situations with guarantees of success. 3.5 Anomalies Detection In some scenarios, the smart home should consider that a wrong decision executed as a consequence of a wrong prediction could be produced due to an anomaly. We can consider this scenario, an inhabitant wake up all working days at 7.00 a.m., so smart home switch on the coffee-maker some minutes before. But when the wake-up alarm goes off, the smart home detects no movement so it could be desirable to request an emergency service with a standard phone call or other mechanism. 3.6 Quick Response When Required Some situations require the smart home prediction algorithms to return a response with time limits. These algorithms are responsible for detection of this situations and they must be able to adapt to these circumstances in order to provide a quick response. 3.7 General Policies and User Adaptation Smart homes technologies should implement the mechanisms to support some home policies like security, energy or comfort. These policies can be collected in different levels. The inhabitants could define some general policies which could be customizable by concrete inhabitant preferences. We can discuss the following scenario; inhabitant A gives preference to energy consumption over comfort. To satisfy these preferences smart home system should try to minimize the energy Perception ReasoningActing Figure 1. The cycle of the automation process in a smart environment. Perception Collector Verifier Repairer Filter Ontologiser Figure 2. Tasks of the perception process. A. Collector This is the lowest-level task, so its main objective is to retrieve data from physical devices. It will usually have to deal with gateways devices of every technology deployed over the smart environment. The properties of the environment suitable to be captured by these sensors are listed in [9]. Actually a subset of these properties could be selected to achieve any particular application or automation process (e.g. put lights off when nobody is at home does not need conditioning information from the environment). Sensors must be deployed in an organized manner to avoid redundant or insignificant information and reduce costs. When developing the Data Collector we have to consider if devices should be programmed or not (or if both types of sensors are present in the environment). Majority devices are pre-programmed and can’t extend its basic functionality (e.g. X-10 Motion sensor), so developers have to adapt to them. On the other hand more and more devices have the ability to extend or modify their performance (e.g. Sentilla Tmote). The latter are much more challenging for developers due to they can be adapted to current needs for a concrete application or circumstances. In this case, Data Collector task covers daemons developed to retrieve information and small applications running on devices as well, so both elements have to agree what information is sent, periodicity of these requests and so on. B. Verifier The main purpose of the Verifier task (as its name means) is to verify that the data (which is currently being received by the Data Collector) is correct. But the challenge now is how the Verifier can determine if the data is correct or not. There is no unique solution for all possible environments, so the verification has to adapt to each environment dynamically. We propose the Verifier to maintain a rule engine where verification rules can be deployed, modified and used in order to determine where data is right or wrong for current environment. The rule engine will have to offer programmers a flexible way to read, add, modify and delete rules. By the way this task has to work side by side with the Repairer when incorrect data is received in order to fix invalid data. The mechanism of communication between the Verifier and the Repairer must be determine. Typical implementations of this mechanism would be the publication of a repair service by the Repairer that will be invoked by the Verifier. We can conclude the Verifier can be seen as a filter applied over all the data received, and it can be used to reject data for any reason (incorrect, redundant and other reasons). C. Repairer The Repairer task will try to fix incorrect data detected by the Verifier. The corrections applied to data can be done in very different ways such us: •Ignore data. First option is to ignore data, it means to set it with unknown value. The Ontologiser will save that value as required by storage system (e.g. Weka-arff ’?’ character, or SQL null value). •Adjust data. If a value is incorrect, but its distance with a correct value is less than a threshold (previously customized) the value could be adjusted to nearest correct value. •Replace data. Other option is to replace data with previous correct data. (e.g. if temperature sensor returns 100 Celsius, and previous data is 25, we could replace current value with 25). •Reject data. If current values are not suitable to be repaired, the repairer will reject it. 398398 Once a set of data has been received, verified and repaired, it has to be sent to the Ontologiser to be organized and stored. D. Ontologiser Artificial intelligence algorithms need a solid base of knowledge to work. The main goal of the Ontologiser is to organize data to conform a model of the real world. This fact demands us a great effort for building a model of a smart environment. Other studies have helped us to compose the model [10] [11] [12] which has been arranged in four main categories explained in [9]. Synchronization.Ontologiser will have to synchronize data from a world full of asynchronous devices and events. Automation applications usually need sets of data composed from multiple devices values. Data from different devices must by synchronized for a latter aggregation, so this task will have to define the rules to synchronize values from multiple and very heterogenous sources. Implementations of the synchronization process will vary depending on the goals of a concrete smart application, but all implementations will share some elements such as: •Data buffer will store data received that is waiting to be paired with other data. •Garbage collector will supervise the size of the buffer, and will periodically clean the buffer from data that couldn’t be paired. Aggregation. Once data is synchronized it must be joined in a set of data for a concrete application. Each attribute should conform the Smart Environment Model [9]. When data is aggregated it is ready to be stored in the knowledge base that feeds the reasoning tasks and learner process. The knowledge base format will be any of the de facto standards like arff, or any relational database. IV. REASONING Reasoning process in smart environments can be separated in several tasks (see figure 3) which interact to achieve three main goals: a) learn, b) reason and c) predict. Learning can be done through any of the techniques commonly known and widely studied in the literature. Attending to the scope of this work we can separate these techniques in two main categories: a) rule based and b)black-box based . Rule based algorithms offers the advantage to be readable and easily understandable by humans. Rule-based techniques are studied in [13] and applied to automation of smart environments in iDorm [14] or combined with temporal reasoning in ADB [15]. On the other hand black-box based techniques don’t offer a comprehensible model of the knowledge that is being learned, but these techniques have been successfully used in pattern recognition and learning in general. Neural Networks or Support Vector Machine are studied as a general technique by authors like [16], [17], [18] and applied to automation of smart environments in the Neural Network House [19], House n [20] and Mav Home [10]. Reasoning Learning Agent Situation Recognition Data Mining Knowledge Representation Configurator Error Detector Figure 3. Typical tasks of the reasoning process. A. Learning Agent This task can be seen as the glue that joins any other task proposed to achieve the objectives of the Reasoning process. It has to deal with: •Data Mining task to extract patterns, •Situation Recognition task to label patterns recognized, •Prediction task to find future actions to be taken and •Error detector task to detect wrong decisions and to make related improvements over the entire process. The learning can be carried out due to the dependence of the user preferences with his habits. For instance normally we have the same lighting preferences in our offices so these preferences must be learned at every environment, due to its dependence with the location, orientation of the window, devices setup and so on. Learning techniques. Two families of algorithms, related with learning machine, can be considered although both are going to be supervised. Support vector machine (SVM) and Neural networks (NN) are some of the techniques we can use to learn user lighting preferences. The main advantage of these techniques is that always offer an output. On the other hand it is hard to understand their outputs which is an important feature that prediction algorithms should implement as expounded in [9]. The other family of algorithms considered are the machine learning techniques based on rules. These algorithms provide 399399 X1 X2All lights on (1,1) Canonical state (0,0) (1,0) (0,1) S1 Figure 4. States for variables x1and x2 an output easier to understand and interpret, but their main disadvantage is that if no rule matches current state, these algorithms don’t offer an output. B. Data Mining Data mining task works side by side with the learning agent to extract information and find patterns from the data stored by the Perception (see section III) process. Smart Environment Vectorization Data mining is fed with a stream of data with a concrete format. This is usually so-called the vectorization of data. This fact forces every concrete smart application to propose a vectorization of the environment and configure all the reasoning process with it. C. Situation recognition A situation is defined as a set of states which have similar values. This set can be defined by: •a canonical state as a representative state of all the states in the set •a distance function which define how to measure the difference between two states •a threshold which define the maximum distance between the canonical state and the states of the set. •a weight for each element of a state between [0,1]. A weight of zero represent that an element must not be considered. The following example illustrates the definition of a situation. Let x1be the variable which represents the state of a light, and let x2be the variable which represent the state of another light. These variables are discrete, and its range is 0,1which represent the states off and on respectively. The figure 4 shows all the states that these variables can constitute. We would like to define the situation S1where any of the lights is on. The canonical state of S1is {1,1}(all lights on) Table I STATES THAT BELONG A SITUATION. State Canon. State Weights dThresh. ⊂S1 x1x2y1y2w1w2 0011 112 1 no 0111 111 1 yes 1011 111 1 yes 1111 110 1 yes for the variables x1and x2respectively. A typical distance function could be defined as: d([x1,x 2],[y1,y 2],[w1,w 2]) = =(x1−y1)2∗w1+(x2−y2)2∗w2 Where [x1,x 2]represent a state, [y1,y 2]represent the canonical state and [w1,w 2]represent the weights of x1 and x2respectively which should be [1,1] for the situation S1any of the lights is on due to both lights should be considered. The threshold can be set to 1. Table I shows the set of states that belong to the situation. If the distance is less or equal to the threshold, the state belongs to the situation. In other cases we can conclude the state will not belong to the situation. D. Prediction The knowledge acquired during the process of learning of the agent will be useful to make predictions about future actions that should be taken in order to achieve any automatism objective. These actions can be considered as the output of smart algorithms as explained in [21]. Desirable features of prediction algorithms. The features that a smart home system must implement are listed, specially related with the prediction algorithms the systems may have. So we don’t focus on artificial intelligence techniques like the studies of [11], [22], [23] do, but on what are the most important and indispensable features that must be considered to develop prediction algorithms for smart environments which are briefly studied in [9] and just listed below: a) Prediction supported by last events and states, b) Predictable by the inhabitants/users, c) Understandable decisions, d) Wrong decisions detection and related improvement, e) Anomalies detection, f) Quick response when required and g) General policies management. E. Error detector The main purpose of this task is to supervise the actions taken by the inhabitants in order to detect opposite decisions between the machine and them. Therefore it will have to keep last actions taken by the smart environment and compare them with actions taken by the inhabitants. The comparison of these actions is usually a complex operation where lots of elements take part into, and it has 400400 to take into account how many actions are going to be considered to make the comparison, how long these actions are going to be considered, or if actions over different devices could be opposite (e.g. open the window could be opposite to switch off the lights). All these aspects make the detection of wrong decisions much more challenging. V. ACTING Finally, to close the cycle, Smart Environments have to act automatically to achieve a concrete smart application. This is the main purpose of the process of Acting. As shown in figure V the decisions and concrete tasks ordered by the Reasoning process have to pass through three main tasks a) policy manager, b) task scheduler and c) task runner1. Acting Policy Manager Task Scheduler Task Runner Figure 5. Tasks of the acting process. A. Policy Manager The decision of carrying out a task must be supervised by the Policy Manager. A smart environment may have defined policies about energy saving, security or comfort. This manager may implement the software artifacts necessary to define, store and query these policies such as Rei policy language [24] or a more extensive work of [25]. This process must be also suitable to change its policies in real time, so inhabitants preferences may vary (i.e. during vacation periods energy saving and security could be a priority in the smart environment). B. Task Scheduler Once an action has been ordered by the reasoning process, and has passed through the filter of the policy manager it has to be scheduled. The scheduler has to consider that some tasks could have a time limit to be executed, and also has to consider its priority. The implementation will typically implement a queue where action will wait. C. Task Runner The task runner is the last task of the Acting process. It has to deal with low level protocols to send orders to devices. This task will receive a set of orders for a set of devices and it will have to translate them to low-level instructions. This task must be a very light process with almost no-lag so its time xcomplexity should be lineal. VI. CASE STUDY We have already shown the Reference Architecture proposed, and the developing environment in [21] so it’s time to present the applications developed during this work to validate the Architecture. We have focused on developing an implementation of the Perception process (first of the Reference Architecture) which main objective is the perception of data about localization of workers, luminosity, temperature and humidity of an office at the LSI department at the University of Seville. A. Mote Sensor Report This application belongs to the Data Collector task. As it was presented in section III-A this task has the responsibility of the low-level interaction with devices, and it could be distributed between central stations of the Smart Environment and the devices themselves. Mote Sensor Report is a pervasive application developed on Sentilla Work IDE and deployed over Sentilla Tmotes. Sentilla Tmotes implement a JVM called Sentilla Point so it can run Java applications. In order to access the hardware capabilities of the device, Sentilla offers a java library which give you access to the data gathered by the sensors and other elements such us extension port or leds. The java application developed access to sensor data and send it through the Zigbee interface. The sensors conform a mesh network where motes act as repeaters. This type of network makes possible to cover wider areas even though Zigbee protocol has a 10m. radio. When developing this application we realized that luminosity captured by sensors fluctuated if the fluorescent lights of the office were on. The reason of this behavior is that fluorescent bulbs are always turning off and on although it is not noticeable by humans due to its short frequency. However this behavior was a problem, so we had to tackle it and we included a part of the Repairer task (see section III-C) at this point, due to the reparation couldn’t be possible in a later moment so it has to be tackled at this point. The solution is quite simple, instead of retrieving just one value, nvalues are retrieved and the average of them is calculated and then sent to the Sentilla Gateway plugged into the central station. B. Mote Dashboard In order to receive all the information from the Sentilla motes an application has been developed. It shows the 401401 information that is being received from the motes in real time. This application carries out three main tasks: •Data Collector. An execution thread is responsible of managing the reception of data from the motes, using the mote gateway supplied with the Development kit. •Verifier. A simple verifier has been developed and performs simple tests, similar to preconditions, over data received. (e.g. luminosity >=0). •Repairer. When the verifier task detects erroneous data Mote Dashboard choose between two options: –Replace. If previous correct data was received short time before, concrete erroneous values are replaced with previous values. The time difference is customizable. –Reject. On the other hand, if previous correct data was received long time before, current erroneous data is rejected. •Ontologiser. Finally correct data is processed by the ontologiser task. It performs three operations: –Synchronization. It keeps a buffer of data received from every mote. When information from outdoor and indoor motes was taken at approximately same time it is composed by the aggregation task. –Aggregation. Once data is synchronized, it is added to the same instance of the input data. –Store. Finally, Dashboard stores all instances in arff format and SQL database (useful for machine learning tools like Weka [13]). Mote Dashboard has other minor functionalities: •Show information. It shows the information received from the motes in a GUI (Fig. VI-B), and a list of active motes in the mesh network. •Reset. Resets the dashboard GUI. This doesn’t affect stored data, but cleans buffers and GUI from data received. Figure 6. Mote Dashboard GUI. VII. CONCLUSIONS We have shown a General Software Reference Architecture to develop Smart Environment Applications which is very useful for developers of Smart Environments solutions. We have answered questions like: a) How should the applications/software be divided?, b) Which process should be present for these smart solutions?, c) How these process should interact? . Future work has to focus on continue developing smart applications to provide more feedback to the Reference Architecture in order to continue improving it. Another interesting challenge is to consider not only the ability of the smart environment to fit user preferences but using Smart Environment to change behaviors in the individual. The increasingly pressure requirement of sustainability points out that future smart environments will teach the inhabitants to be more conscious with the world and its environment and we believe that it has to be the way future smart environments must work. ACKNOWLEDGMENT This research is partially supported by the MEC I+D project InCare. Ref: TSI2006-13390-C02-02 and the Andalusian Excellence I+D project CUBICO Ref: TIC2141. REFERENCES [1] G. Youngblood, E. Heierman, L. Holder, and D. Cook, “Automation intelligence for the smart environment,” Proceedings of the International Joint Conference on Artificial Intelligence, 2005. [Online]. Available: http: //ijcai.org/papers/post-0192.pdf [2] M. Weiser, “The computer for the 21st century,” SIGMOBILE Mob. Comput. Commun. Rev., vol. 3, no. 3, pp. 3–11, 1999. [Online]. Available: http://www.ece.ubc.ca/∼guenther/ LearningInMulti-AgentControlOfSmartMatter.pdf [3] J. A. Alvarez, M. D. Cruz, A. Fernndez, J. A. Ortega, and J. Torres, “Experiencias en entornos de computacin ubicua mediante arquitecturas orientadas a servicios,” CEDI-JSWeb, pp. 167–174, sept 2005. [4] D. Cook and S. Das, “How smart are our environments? An updated look at the state of the art,” Pervasive and Mobile Computing, vol. 3, no. 2, pp. 53–73, 2007. [Online]. Available: http://www.sciencedirect.com/science? ob=ArticleURL& udi=B7MF1-4MP002G-1& user= 603129& rdoc=1& fmt=& orig=search& sort=d&view= c& acct=C000031118& version=1& urlVersion=0& userid=603129&md5=560f54d55b3fe9bdeb9c64f72b45e751 [5] C. da Costa, A. Yamin, and C. Geyer, “Toward a general software infrastructure for ubiquitous computing,” Pervasive Computing, IEEE, vol. 7, no. 1, pp. 64–73, Jan.-March 2008. [6] T. Weis, M. Knoll, A. Ulbrich, G. Muhl, and A. Brandle, “Rapid prototyping for pervasive applications,” Pervasive Computing, IEEE, vol. 6, no. 2, pp. 76–84, April-June 2007. [7] K. Rehman, F. Stajano, and G. Coulouris, “An architecture for interactive context-aware applications,” Pervasive Computing, IEEE, vol. 6, no. 1, pp. 73–80, Jan.-March 2007. 402402 [8] D. Bannach, P. Lukowicz, and O. Amft, “Rapid prototyping of activity recognition applications,” Pervasive Computing, IEEE, vol. 7, no. 2, pp. 22–31, April-June 2008. [9] A. Fern´ andez-Montes, J. ´ Alvarez, J. Ortega, M. Cruz, L. Gonz´ alez, and F. Velasco, “Modeling Smart Homes for Prediction Algorithms,” Lecture Notes in Computer Science, vol. 4693, p. 26, 2007. [Online]. Available: http://www. springerlink.com/content/m6q1051w442nv845/fulltext.pdf [10] D. Cook, M. Youngblood, and S. Das, “A multi-agent approach to controlling a smart environment,” Lecture notes in computer science, vol. 4008, p. 165, 2006. [Online]. Available: http://www.springerlink.com/content/ 142027w275531607/fulltext.pdf [11] S. Das and D. Cook, “Designing Smart Environments: A Paradigm Based on Learning and Prediction,” Mobile, Wireless, and Sensor Networks. [Online]. Available: http://www. springerlink.com/content/7t87564r42285713/fulltext.pdf [12] J. Li, Y. Bu, S. Chen, X. Tao, and J. Lu, “FollowMe: On Research of Pluggable Infrastructure for Context-Awareness,” Proceedings of the 20th International Conference on Advanced Information Networking and Applications-Volume 1 (AINA’06)-Volume 01, pp. 199–204, 2006. [Online]. Available: http://doi.ieeecomputersociety.org/10.1109/AINA. 2006.182 [13] I. Witten and E. Frank, Data Mining: Practical Machine Learning Tools and Techniques. Morgan Kaufmann, 2005. [14] F. Doctor, H. Hagras, and V. Callaghan, “A fuzzy embedded agent-based approach for realizing ambient intelligence in intelligent inhabited environments,” Systems, Man and Cybernetics, Part A, IEEE Transactions on, vol. 35, no. 1, pp. 55–65, Jan. 2005. [15] J. Augusto and C. Nugent, “The use of temporal reasoning and management of complex events in smart homes,” Proceedings of European Conference on Artificial Intelligence (ECAI 2004), 2004. [Online]. Available: http: //www.infj.ulst.ac.uk/∼jcaug/ecai2004.pdf [16] C. Bishop, Neural Networks for Pattern Recognition. Oxford University Press, USA, 1995. [17] D. MacKay, Information Theory, Inference and Learning Algorithms. Cambridge University Press, 2003. [Online]. Available: http://www.inference.phy.cam.ac.uk/itprnn/book. pdf [18] N. Cristianini and J. Shawe-Taylor, An Introduction to Support Vector Machines. Cambridge University Press, 2000. [Online]. Available: http://books.google.es/books? hl=es&lr=&id=L-2Vqx56J5UC&oi=fnd&pg=PR9&dq=an+ introduction+to+support+vector+machines+cristianini&ots= eszbtwZefz&sig=TiSbqVWHmNNZ-WS-OmReCn9soZI [19] M. Mozer, “Lessons from an adaptive home,” Smart Environments: Technology, Protocols, and Applications, pp. 273–298, 2005. [Online]. Available: http://www.cs.colorado. edu/∼mozer/papers/reprints/smart environments.pdf [20] K. Larson, “House n Living Laboratory White Paper,” 2001. [21] A. Fernandez-Montes, J. Ortega, L. Gonzalez, J. Alvarez, and M. Cruz, “Smart Environment Vectorization: An Approach to Learning of User Lighting Preferences,” Lecture Notes in Computer Science, vol. 5177, pp. 765–772, 2008. [22] H. Hagras, V. Callaghan, M. Colley, G. Clarke, A. Pounds-Cornish, and H. Duman, “Creating an Ambient- Intelligence Environment Using Embedded Agents,” 2004. [Online]. Available: http://doi.ieeecomputersociety.org/10. 1109/MIS.2004.61 [23] N. Roy, A. Roy, and S. Das, “Context-Aware Resource Management in Multi-Inhabitant Smart Homes: A Nash H- Learning based Approach,” Proc. of 4th IEEE Int’l Conf. on Pervasive Computing and Communications (PerCom2006), 2006. [Online]. Available: http://doi.ieeecomputersociety.org/ 10.1109/PERCOM.2006.18 [24] L. Kagal, T. Finin, and A. Joshi, “A policy language for a pervasive computing environment,” Policies for Distributed Systems and Networks, 2003. Proceedings. POLICY 2003. IEEE 4th International Workshop on, pp. 63–74, June 2003. [Online]. Available: http://ieeexplore.ieee.org/iel5/8577/ 27164/01206958.pdf?tp=&isnumber=&arnumber=1206958 [25] J. Ahn, B. Chang, and K. Doh, “A Policy Description Language for Context-based Access Control and Adaptation in Ubiquitous Environment,” TRUST06, August, 2006. [Online]. Available: http://www.springerlink.com/content/ 67g537107314j573/fulltext.pdf 403403 Paulo Novais, Davy Preuveneers, and Juan M. Corchado (Eds.) Ambient Intelligence – Software and Applications 2nd International Symposium on Ambient Intelligence (ISAmI 2011) ABC Editors Prof. Paulo Novais Universidade do Minho Departamento de Informática Campus de Gualtar 4710-057 Braga Portugal Dr. Davy Preuveneers Katholieke Universiteit Leuven Department of Computer Science Celestijnenlaan 200A, 02.152 B-3001 Heverlee Belgium Prof. Juan M. Corchado Universidad de Salamanca Departamento de Informática y Automática Facultad de Ciencias Plaza de la Merced S/N 37008, Salamanca Spain E-mail: [email protected] ISBN 978-3-642-19936-3 e-ISBN 978-3-642-19937-0 DOI 10.1007/978-3-642-19937-0 Advances in Intelligent and Soft Computing ISSN 1867-5662 Library of Congress Control Number: 2011923221 c 2011 Springer-Verlag Berlin Heidelberg This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable for prosecution under the German Copyright Law. The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Typeset & Cover Design: Scientific Publishing Services Pvt. Ltd., Chennai, India. Printed on acid-free paper 543210 springer.com Contents XI Mobile Personal Health Systems for Patient Self-management: On Pervasive Information Logging and Sharing within Social Networks .............................. 141 Andreas K. Triantafyllidis, Vassilis G. Koutkias, Ioanna Chouvarda, Nicos Maglaveras Multi-Agent Strategy Synthesis in Smart Meeting Environments ................................................ 149 Ren´eLeistikow Reusable Gestures for Interacting with Ambient Displays in Unfamiliar Environments ................................. 157 Radu-Daniel Vatavu Graphs and Patterns for Context-Awareness................. 165 Andrei Olaru, Adina Magda Florea, Amal El Fallah Seghrouchni ARTIZT: Applying Ambient Intelligence to a Museum Guide Scenario .............................................. 173 Oscar Garc´ıa, Ricardo S. Alonso, Fabio Guevara, David Sancho, Miguel S´anchez, Javier Bajo Workshop on the Reliability of Intelligent Environments (WoRIE) Reliability of Location Detection in Intelligent Environments ................................................ 181 Shumei Zhang, Paul J. McCullagh, Chris Nugent, Huiru Zheng, Norman Black Flexible Simulation of Ubiquitous Computing Environments ................................................ 189 Francisco Campuzano, Teresa Garcia-Valverde, Alberto Garcia-Sola, Juan A. Botia Safety Considerations in the Development of Intelligent Environments ................................................ 197 Juan Carlos Augusto, Paul J. McCullagh LECOMP: Low Energy COnsumption Mesh Protocol in WSN ........................................................ 205 Juan A. Ortega, Alejandro Fernandez-Montes, Daniel Fuentes, Luis Gonzalez-Abril 210 J.A. Ortega et al. As batteries drain the Hserver computes new rules, and periodically, the Hserver reconfigures the routing rules to balance batteries of the whole network. In addition to this reconfiguration if a node stops working due to an attack, the lack of battery or another external factor it will not send any reply. In this case, the Hserver will check affected paths and compute new paths to replace invalid ones. 3 Implementation, Tests and Results For test purposes we have implemented the routing proposal in Sentilla Work, the IDE supplied with Sentilla Development Kit. The real implementation is an extension of one of the protocols supplied by Sentilla, therefore the algorithm shown it was written in Java. Results are shown in Table 1 below where the base mesh protocol,the first version of the method called LECOMP-1 [1] and present method LECOMP-2 are compared. Tests have been done at the WSN deployed at Computer and Systems Languages department of the University of Sevilla. It consists on 9 Zigbee sensors deployed in 9 offices. The number of avoided messages in LECOMP-1 and LECOMP-2 is calculated through an approximation between the number the time computed and the duplicated messsages received in the base case. Table 1 shows that in case of LECOMP-1 the sensor network lifetime is 421h and for LECOMP-2 439h. Compared to 360h when no LECOMP is applied this is an improvementof approximately 17% in case LECOMP-1 and 22% of LECOMP-2 in terms of sensor network lifetime. That implies that when LECOMP-1 is used the WSN runs 2,5 days longer and in case of LECOMP-2 over 3 days longer than when no intelligent routing applied. Moreover, we would like to emphasize the improvement of about 7% in avoided messages and 10% in avoided steps in this proposal comparing with the first version of the method. In figure 3, the battery life and the (no-duplicated) received messages Table 1 Base-Protocol, LECOMP-1 and LECOMP-2 comparison Base LECOMP-1 LECOMP-2 #sensors 9 9 9 timeComputed 360h. 421h. 439h. #msgsPerHour 30 29.5 29.7 #msgs 10800 12425 13044 trainingTime 0h. 1h. 1h. #duplicatedmsgs 1944 5 5 avgStepsPerMsg 2,3 2,02 2,12 #msgsReceived 12744 12430 13049 #msgsAvoided 02249 2945 %avoidedMsgs 018,1 22,5 #Steps 29311 25524 27933 #StepsAvoided 07368 9020 %avoidedSteps 025,1 35,9 LECOMP: Low Energy COnsumption Mesh Protocol in WSN 211 in the three cases are compared. Using our method the number of received messages and the lifetime of the WSN increments. Figure 3 the battery life and the (no-duplicated) received messages in the three cases is compared. Using our method the number of received messages and the lifetime of the WSN increments. 60 70 80 90 100 ) 10 20 30 40 50 60 Battery Level (% Base LECOMP-1 LECOMP-2 0 0 5000 10000 15000 Received Messages Fig. 3 Comparison Graphic between Base Mesh protocol, LECOMP V1 and LECOMP V2 4Conclusion In this paper, we have presented a energy-efficient method for mesh wireless sensor networks. We have got to save up energy by reducing the forwarding of messages. First, the central server determinesthe rules throughthe informationof the messages routes, the battery levels and the distance between sensors. Second, the network is reconfigured avoiding duplicated messages and balancing the load. The described practical experiences with over Sentilla sensors and Zigbee technology shows an improvement in the WSN lifetime of 22% compared with the base protocol. Acknowledgements. This research is supported by the MCI I+D project ARTEMISA (TIN2009-14378-C02-01). References 1. Fuentes, D., Fernandez-Montes, A., Ortega, J.A., Gonzalez-Abril, L.: Energy-Efficient Routing Approach for Wireless Sensor Networks. In: Proceedings of the 4th Symposium of Ubiquitous Computing and Ambient Intelligence UCAmI 2010, pp. 9–12 (2010) 2. Shrestha, A., Xing, L.: A Performance Comparison of Different Topologies for Wireless Sensor Netwoks. In: IEEE Conference on Technologies for Homeland Security 2007, pp. 280–285 (2007) 212 J.A. Ortega et al. 3. Chandrakasan, A., Min, R., Bhardwaj, M., Cho, S.H., Wang, A.: Power Aware Wireless Microsensor Systems. In: Keynote Paper. 28th European Solid-State Circuits Conference (ESSCIRC), pp. 47–54 (2002) 4. Anastasi, G., Conti, M., Di Francesco, M., Passarella, A.: Energy conservation in wireless sensor networks: A survey. In: Proceedings of the 33rd Hawaii International Conference on System HICSS 2000, pp. 10–20 (2000) 5. Mathur, G., Desnoyers, P., Chukiu, P., Ganesan, D., Shenoy, P.: Ultra-low power data storage for sensor networks. In: Proceedings of the 5th International Conference on Information Processing in Sensor Networks, pp. 374–381 (2005) 6. Shih, E., Cho, S., Ickes, N., Min, R., Sinha, A., Wang, A., Chandrakasan, A.: Physical Layer Driven Protocol and Algorithm Design for EnergyEfficient Wireless Sensor Networks. In: Proceedings of the 7th Annual International Conference on Mobile Computing and Networking, pp. 272–287 (2001) 7. Pantazis, N.A., Nikolidakis, S.A., Vergados, D.D.: Energy-Efficient Routing Protocols in Wireless Sensor Networks for Health Communication Systems. In: Proceedings of the 2nd International Conference on PErvsive Technologies Related to Assistive Environments, vol. 34 (2009) 8. Zeng, K., Ren, K., Lou, W., Moran, P.J.: Energy Aware Efficient Geographic Routing in Lossy Wireless Sensor Networks with Environmental Energy Supply. In: Proceedings of QShine 2006 (2006) 9. Chen, X., Blum, R., Sadler, B.M.: A New Scheme for Energy-efficient Estimation in a Sensor Network. In: Proceedings of the Conference on Information Sciences and Systems, CISS (2009) 10. Marcelloni, F., Vecchio, M.: A simple algorithm for data compression in wireless sensor networks. IEEE Communications Letters 12(6) (2008) 11. Tang, Q., Yang, L., Giannakis, G.B., Qin, T.: Battery Power Efficiency of PPM and FSK in Wireless Sensor Networks. IEEE Trans. on Communications 6(4), 1308–1319 (2007) 12. Hart, P.E., Nilsson, N.J., Raphael, B.: A Formal Basis for the Heuristic Determination of Minimum Cost Paths. IEEE Transactions on Systems Science and Cybernetics 4(2), 100–107 (1968) PART IV Final remarks Page 147 CHAPTER 12 CONCLUSIONS AND FUTURE WORK The law of conservation of energy tells us we can't get something for nothing, but we refuse to believe it. Isaac Asimov, 1988 12.1 Conclusions This thesis focuses on the problem of the growing energy consumption worldwide, which presents major diculties in the short and medium-term. This obstacle has two bases: on one hand, economic costs of energy, which can prevent companies and populations from developing. On the other hand, it is known that energy sources are limited, and that the current growth of energy consumption remains not sustainable. Page 149 Chapter 12: Conclusions and future work Moreover, this work analyzes both energy consumption in Grid-computing and Smart environments, and provides solutions to save the maximum energy. It has been proved that Grid facilities can save up to 40% of energy and that up to 72% of energy in lighting conditioning can be saved by making use of proper policies. Furthermore, the cost of applying these energy-saving policies is insignicant in comparison to the saving in costs involved with less energy consumption. Simulation software has been developed in order to compute energy consumption at Grid locations, whereby several energy-saving and aggregation policies have been applied. This tool has been widely tested in order to obtain reliable results, and has been published online through the Idinfor web page, thereby making it available worldwide for researchers interested on these issues. Finally, the innovation of the thesis is also presented in the use of Data Envelopment Analysis, which is applied in order to compare the eciency between various Grid-Computing locations. This innovation has proved itself to be a useful tool for Grid-system administrators in the evaluation of eciency and performance of locations, who in turn recommend actions that should be carried out for the location to become as ecient as the most ecient locations. Location eciency varies from 57% to 68% in comparison with the most ecient locations (100%). These non- ecient locations are encouraged to decrease their number of resources and increase the number of jobs run. 12.2 Future work In terms of energy saving in Grid-Computing and Smart Environments, future work will focus on six topics:  Adapting energy policies to take into account the variety of energy consumption of resources. To date, every resource has been assumed to consume the same amount of energy. However energy consumption may dier from one Page 150 12.2 Future work resource to another. This fact can make slight dierences between energy policies and an arranging policy that attempts to maximize the usage of more ecient resources could provide better results.  Adapting models to issues of data centers. Here, we have provided solutions for Grid-Computing Environments, where jobs are deployed for periods from minutes to hours, and deployed to resources numbering from two to several. Data centers usually process atomic queries by means of a single resource. For example, Google data centers have to process web searches which usually are resolved within one second.  Creation of new policies. With respect to adapting models to issues of data centers, policies should be adapted to this new model in meeting objectives for the saving of energy. Moreover policies analyzed in this thesis work can also be combined and the results of these combinations can be analyzed. For example, we could establish a new policy combining Load policy and Exponential policy . In this way, if the current load is low, then Load policy would shut resources down, and subsequently Exponential policy could be applied in order to decide what to do with the resources.  Exporting techniques to other supercomputing infrastructures. Since 2009 the Centro Informático Cientíco de Andalucía (CICA) cluster for supercomputing has been logging information about jobs, deployments and usages, and hence we have started to apply these energy-saving policies to simulate this cluster in order to test how much energy could be saved, and to check whether the performances of these policies would present similar behaviour in other supercomputing environments.  Within the context of Smart Environments, we are planning to share our experience and knowledge with the University of Reutlingen and the HTWG Konstanz in order to build a new model of Smart Environment where biometric information on inhabitants is considered, such as the level of stress. Page 151 APPENDIX A CURRICULUM A.1 Published papers During last years, a set of research papers has been developed, within both Grid and Smart Environments and most of them related to energy saving. [ 2007 ] : During 2007 the research was focused on modeling smart environments and oering applications which make use of the control of these environments. We also applied for a national researching project called InCare to improve services for dependant people. [ 2008 ] : In 2008, we focused our eorts in the development of National project InCare and Regional project CUBICO. We setup a Smart Environment lab, where we could learn user preferences in terms of comfort and security. We collaborated with University Carlos III de Madrid, where the PhD candidate made an stage for one month, and we published a research paper in common [49] and a book chapter [73]. [ 2009 ] : This year we decided to focus on Saving Energy and Environmental sustainability within the context of Smart Environments. Papers [50, 24] and others Page 153