In e na ional Jou nal o Compu e Science & In o ma ion Technology (IJCSIT) Vol 10, No 4, Augus 2018.
DOI: 10.5121/ijcsi .2018.10403 25
G
ROUP
B
ASED
R
ESOURCE
M
ANAGEMENT
A
ND
P
RICING
M
ODEL
I
N
C
LOUD
C
OMPUTING
Shelia Rahman
1,
A oza Sul ana
2
, A sana Islam
3,
and Md Whaiduzzaman
4
1
Ins i u e O In o ma ion Technology,Jahangi naga Uni e si y,Dhaka,Bangladesh.
2
Ins i u e O In o ma ion Technology,Jahangi naga Uni e si y,Dhaka,Bangladesh.
3
Ins i u e O In o ma ion Technology,Jahangi naga Uni e si y,Dhaka,Bangladesh.
4
Ins i u e O In o ma ion Technology,Jahangi naga Uni e si y,Dhaka,Bangladesh.
A
BSTRACT
Cloud compu ing u ilizes la ge scale compu ing in as uc u e ha has been adically changing he IT
landscape enabling emo e access o compu ing esou ces wi h low se ice cos , high scalabili y ,
a ailabili y and accessibili y. Se ing asks om mul iple use s whe e he asks a e o di e en
cha ac e is ics wi h a ia ion in he equi emen o compu ing powe may cause unde o o e u iliza ion
o esou ces.The e o e main aining such mega-scale da acen e equi es e icien esou ce managemen
p ocedu e o inc ease esou ce u iliza ion. Howe e , while main aining e iciency in se ice p o isioning i
is necessa y o ensu e he maximiza ion o p o i o he cloud p o ide s. Mos o he cu en esea ch
wo ks aims a how p o ide s can o e e icien se ice p o isioning o he use and imp o ing sys em
pe o mance. The e a e compa a i ely ewe speci ic wo ks ega ding esou ce managemen which also
deals wi h he economic sec ion ha conside s p o i maximiza ion o he p o ide . In his pape we
ep esen a model ha deals wi h bo h e icien esou ce u iliza ion and p icing o he esou ces. The join
esou ce managemen model combines he wo k o use assignmen , ask scheduling and load balancing
on he ac o CPU powe endo semen . We p opose ou algo i hms espec i ely o use assignmen , ask
scheduling, load balancing and p icing ha wo ks on g oup based esou ces o e ing educ ion in ask
execu ion ime(56.3%),ac i a ed physical machines(41.44%),p o isioning cos (23%) . The cos is
calcula ed o e a ime in e al in ol ing he numbe o se ed cus ome a his ime and he amoun o
esou ces used wi hin his ime.
K
EYWORDS
Resou ce Managemen , Resou ce P icing, Task Execu ion, Load Balancing, Task Scheduling.
1.
I
NTRODUCTION
Cloud compu ing is a pool o i ual machines wi h unde lying da acen e s o physical machines
p o iding a ious kinds o agile and e ec i e se ices o he use in a o m o i ualiza ion
o e e y kind o compu ing se ices om in as uc u e o so wa e[1] [2].Wi h he de elopmen
inIn e ne uses o In e ne enabled de ices a e inc easing day by day in ac he numbe o
IoTenabled de ices ha e al eady ou numbe ed o al human popula ion[3]. This esul s in mo e
gene a ion o la ge scale da a which equi es as e p ocessing and as e ask
esponse.The e o emo e de ices a e now ge ing connec ed o he cloud as hese de ices a e
limi ed o s o age andp ocessing powe .Howe e wi h he inc ease in use s cloud se ice
p o ide s a e now mo e complied o use la ge and powe ul da acen e s as in es men s we e
done by many elecommunica ion companies in o de o sa is y hei g owing cus ome
equi emen s and a oiding any SLA iola ion[4].These da acen e s a e equipped wi h powe ul
ha dwa e and connec ed wi h highbandwid h ne wo ksand managed wi h so wa e esou ces. . As
a esul he equi emen o e icien managemen p ocedu es o handling such la ge da acen e s
a e becoming a opic o in e es o he esea che s.Cu en esea ch wo ks mos ly aims a be e
se ice p o isioning o he use and imp o ing sys em pe o mance wi h only some speci ic wo ks
In e na ional Jou nal o Compu e Science & In o ma ion Technology (IJCSIT) Vol 10, No 4, Augus 2018.
26
ega ding economic and cos ing model.These models deal wi h he esou ce managemen in cloud
compu ing which assu ese icien esou ce managemen ha ensu es maximiza ion o p o i o
he p o ide .As Bo h use and p o ide s play impo an ole in shaping o cloud compu ing wi h
clea and dis inc p omising ea u es , esou ce managemen p ocedu e ha holds he end o he
ba gain o bo h use s andp o ide s a e impo an .
Figu e 1: Cos based p ice managemen in cloud compu ing
Resou ce managemen in cloud co e s a numbe o issues such as e icien ask scheduling,
esou ce p o isioning, esou ce alloca ion, esou ce adap ion, asks assignmen ,load balancing and
o he s.Among hem h ee impo an ea u es o esou ce managemen can be conside ed esou ce
scheduling,load balancing and ask assignmen o shape he esou ce u iliza ion in cloud
en i onmen . Resou ce scheduling e e s o a ime able o e en s main aining e en s and
esou ces o de e mine he ime an ac i i y is assigned o execu ion.Load balancing aims o e enly
dis ibu ing asks o e physical machines o educe a ic in any pa icula ones and o maximize
he numbe o machines in s and by mode o educe powe usage. Tasks assignmen aims o
mapping asks o aVM and gene ally chooses he VM whe e i can execu e as e .The
p o isioning and placemen o VMs mus be done e icien ly aking in o accoun he a ailable
esou ces in cloud .Again, he econ igu a ions mus be pe o med as o esize o elease he
exis ing i ual esou ces because o he a iabili y and elas ici y o esou ce demand[5].
Ine icien esou ce managemen nega i elya ec s pe o mance and cos as well as impai ing
sys em unc ionali y.
Ensu ing P o i o e e y cloud p o ide is a undamen al goal.T adi ional p ocedu es such as
sys em op imiza ion ends o aim a di e en sys em pe o mance me ics based on sys em
pa ame e s and cons ain s a he han economic ac o s( he p o i ,cos ,and e enue)[5].A cos
based modelcalcula es he p ice o a se ice using allo e cos o p o iding he se ice as well as
adding ape cen age o he cos as desi ed p o i [5][6].This model uses 1. ixed cos and 2. a iable
cos ocalcula e he o al cos [5].Fixed cos includes cos equi ed o se ing up ha dwa e, se e s
andne wo k de ices, equi ed human esou ces. On he o he hand a iable cos is no cons an ,
i a ies acco ding o he numbe o sales p oduced by he se ice such as ene gy,
bandwid h,cos o ans e ing da a be ween di e en da acen e s and so on.P o ide s use cos
based p icing o la ge scale da acen e s o calcula e he cos o a se ice o he incoming use s
asks eques s[7].
In e na ional Jou nal o Compu e Science & In o ma ion Technology (IJCSIT) Vol 10, No 4, Augus 2018.
27
Resou ces in a cloud se e includes bo h ha dwa es and so wa e. Cloud compu ing
in as uc u e is gene ally hos ed on da acen e s and hese da acen e s can be geog aphically
dis ibu ed and connec ed using high speed ne wo k.A hype iso so wa e emula es se e ’s
CPU,memo y ,ne wo kand o he esou ces o enable isola ion o mul iple use s when unning on
he same physical machine.Tasks gene a ed om use de ices a e assigned o i ual machines
wi h cloud b oke . Taskshandled by a se e a e o hyb id in na u e whe e some asks equi e
mo e compu a ion powe hano he s and again hey can di e in p io i ies.IaaS-Clouds o e VMs
a a speci ic mone a y cos .Excess consump ion o i ual esou ces need o be a oided as
unde u ilized wo ke -nodes educe he mone a y alue he cloud clien achie es.The numbe o
se ing VMs mus ne e su pass he h eshold o e which excess VMs become an o e head o
powe consump ion.The e o e assigning asks o VMs o ensu e bo h as e execu ion and be e
u iliza ion o esou ces a he same ime is a icky issue as du ing se ice p o isioning he
p o ide ha e o main ain bo h Quali y o Se ice(QoS) and gaining he maximum e enue.
In his pape we ha e iden i ied e icien esou ce managemen wi h p o i maximiza ion aises
h ee impo an issues-1.Wha ypes o asks a e needed o assign 2.How da acen e s should
place he asks o execu ion on VMs, 3.How o educe he numbe o ac i e se e s a a
da acen e along wi h se ing maximum numbe s o use s,4.How he use s a e p iced so ha
p o ide s canha e maximized p o i . In his esea ch we answe he ques ions by modeling a
esou ce managemen sys em and use cos based p icing model o calcula e he p ice o esou ces
used.In ou p oposed model we logically di ide he compu ing esou ces in a da acen e in o
g oups called Resou ceG oup based on wha ype o asks will be assigned o hem .A numbe o
VMs a e assigned o each g oup whe e VMs wi hin same g oup a e assigned o execu e asks o
equi ing simila amoun o compu ing powe . Tasks assignmen and scheduling wo ks in a
sequen ial manne .Fou di e en algo i hms a e p esen ed which wo ks o achie e a o e all p o i
o he p o ide and as e ask execu ion o he use .
In he es pa o his a icle we ha e desc ibed some scena ios o esou ce managemen
andp icing in cloud in Sec ion 2. A e ha we ha e ep esen ed ou p oposed sys em
a chi ec u ein Sec ion 3 and algo i hms o ask assignmen , ask scheduling and load balancing in
Sec ion 4.La e we ha e e alua ed ou p oposed me hod wi h some o he exis ing ones p esen ed
in sec ion5.
2.
R
ELATED
W
ORKS
The e a e se e al wo ks ega ding esou ce managemen in cloud compu ing. Di e en cos ing
models ha e been p oposed o p icing he se ices. Howe e he e a e e y ew esea ch which
collabo a es cos managemen wi h esou ce managemen in cloud compu ing. The managemen
se ice obse es mechanisms in he i ual se e s wi hin he cloud compu ing sessions which
allows i omoni o ,analyse as well as p o ide ep o s along wi h ale ing pe aining o
pe o mance me icso he a ious i ual se e s[8].In [9] au ho s p oposed an ene gy e icien
adap i e esou ce managemen o ehicula cloud o maximize he o e all communica ion and
and compu ing ene gye iciency.I mee s he applica ion-induced ha d QoS equi emen s wi h
leas ansmission a es bu wi h maximum delays and delay-ji e s.In [10] au ho ga e hei
conce n on ne wo k isualiza iono esou ce alloca ion dynamically in cloud compu ing and he
impo ance o mee ing QoS. When he bes e o alls hen he Se ice Le el Ag eemen (SLA)
akes he equi alen ac ions.Besidesdi e en scheduling and load balancing algo i hm has been
p oposed [11][12][13][14].In [15] esea che s p oposed o schedule he asks based on
equi emen o compu a ion powe and [14] ep esen ed a load balancing mechanism also based
on powe equi emen .Au ho s in [16] p oposed a esou ce u iliza ion me hod based on g eedy
me hod.Acco ding o [17][18][19] when using mobilede ices augmen ed wi h cloudle s o
capaci y inc emen he o e ed se ice o load a io is needed o be conside ed. . These esea ch
wo ks conside s only limi ed cons ain s a he han a comple e scena io o esou ce managemen .
In e na ional Jou nal o Compu e Science & In o ma ion Technology (IJCSIT) Vol 10, No 4, Augus 2018.
28
Besides how p o iding esou ce e iciency a ec s p o ide s e enue is no discussed which
a ec s in long e m p o isioning.
Va ious model o p icing cloud se ice ha e been p oposed o cloud ma ke scena io e.g
cos based p icing [7][20][21] depic s how he cos o he esou ces used can be calcula ed,p o i
maximiza ion [22][23] aiming o inc easing p o i o he p o ide , di e en ial p icing [24][15]
cha gesdi e en p ice o di e en clien based on hei demand. Acco ding o [2] cha ges o he
se ice eques a e decided by he cos ing mechanism depending on he ime o submission, a e o
p icingo he a ailabili y o esou ce and accoun ing mechanism is used o calcula ion o he
ac ual usageo esou ces.A de ailed o e iew o cloud esou ce managemen using economic
analysis is gi enin [5].Techno-economic modelling is used in [20] o assess he cos e iciency o
using SDN in o he LTE ne wo k.Acco ding o [21] Pa icula ly o mos demand side
managemen applica ionsbecause o he a iabili y o cus ome beha iou s2, he compu ing
equi emen s luc ua es signi ican ly which discou ages p o ide s o se ing up new
da acen e s.The seemingly ne e -ending esou ces o a physical cloud, exac ing he speci ic
amoun o esou ces equi ed mus be based onbo h he consume ’s in ended amoun o
expendi u e and he pe o mance bo lenecks which can be isible only a un ime[22].They
p opose o con inuously moni o ing use applica ion pe o manceand emo ing o adding VMs
when pe o mance luc ua ions is obse ed in se ing he needs o au onomic sys ems. To
maximize he p o i , a cloud p o ide needs o unde s and bo h se icecha ges and business cos s
as well as he way hey a e by he cha ac e is ics o he applica ionsand he con igu a ion o a
mul i-se e sys em[23].P icing also di e s based on he geo-dis ibu iono da acen e s.A
dynamic p icing along wi h p o i maximiza ion aiming p icing in geog aphicallydis ibu ed
da acen e s in cloud is p oposed in [25].
3.R
ESOURCE
G
ROUP
M
ODELING
In his esea ch wo k we p opose ha he esou ces a e di ided in o g oups. Acco ding o wha
se ices he p o ide is o e ing he g oups a e c ea ed based on he equi emen o bo h he
p o ide and he use .Fo example le s hink a cloud p o ide has N geog aphically dis ibu ed
da acen e s.Based on he job ype asks a e di ided in h ee g oups whe e hey di e s in hei
equi emen o compu a ion powe .Based on his he da acen e esou ces can be di ided in o
h ee g oups. Each g oups con ains a numbe o VMs and he g oups a e u he di ided in o
subg oups. Ou p oposed model pa i ions all he se e esou ces based on he CPU powe
equi ed which implies he numbe o CPU cycles and he amoun o in e nal memo y he g oups
a e assigned. The g ouping is done logically which implies ha hey don’ necessa ily ha e o be
physically sepa a ed.
3.1 Resou ce G oup
A Resou ceG oup is like a logical con aine ha ha e ce ain amoun o compu ing esou ce
assigned o hem. P o ide can deploy, manage and moni o all he esou ces o p o isioning
solu ion as a g oup, a he han handling hese esou ces indi idually. The g oups can be di ided
in o subg oups when equi ed. A sequen ial app oach is ollowed in deploying and balancing
he esou ces.La ge g oup o esou ces a e again di ided in o subg oups. G oups a e di e en
wi hone ano he on he basis o amoun o esou ce consump ion.Le s he p o ide ’s esou ces in
ada acen e is di ided in o wo g oups A and B.G oup A is o se icing use s demanding a
la geamoun o powe and he o he one is o he less demanding. The i s g oup o e s ins ances
ha is cons i u ed o mo e numbe s o CPU and mo e Powe ul (Speed, MIPS) CPU and RAM
powe hen he second g oup. As he la ge g oups a e spli in o smalle subg oups con ol
becomes mo edis ibu ed. The assignmen ,scheduling, load balancing algo i hms wo ks
di e en ly on his subg oups and synch onizes when equi es. The subg oups con ains a ce ain
numbe o VMs. EachVM is assigned o use asks acco ding o hei demand. An a ay is used o
con ain iden i ica ionand s a us o his VMs a a ce ain ime. Wi hin his a ay he VMs o e s a
In e na ional Jou nal o Compu e Science & In o ma ion Technology (IJCSIT) Vol 10, No 4, Augus 2018.
29
sequen ial ela ion and ela i i y among VMs ha belongs o he same g oup. The esou ces in
each subg oup is o e edas ins ances. Fo simpli ying ou desc ip ion each ins ance will be
add essed as a PM. Resou cesa e sha able among he g oups. Bu he sha ing should be
minimized. A da acen e has Resou ceG oup iden i ied as A, B, C. . . .N whe e N= o al numbe o
Gene al G oup and h eshold o h1, h2, h3.... h1 .Then incoming asks can be g ouped based on
he h eshold such as asks equi ing powe below h1 a e assigned o g oup A , asks be ween h1
and h2 a e submi ed o g oup Band goes on.Tasks in he same g oup need simila amoun o
compu a ion powe bu la gely di e swi h o he g oups bu p e y simila wi hin he same g oup.
Subg oups a e iden i ied as ollowsand asks in he subg oup a e simila in equi emen o
powe .A pa icula numbe o VMs a eassigned o a subg oups , du ing ask mo ing o load
balancing o emapping VMs wi hin sameg oups a e p io i ized. Subg oup wi hin each g oup= 0,
1, 2. . . . . . n
Se ice poin (VM) in g oup:
G oup A – subg oup 0 VM=0, 1, 2. . . . . . n
G oup A – subg oup 1 VM=n+1, n+2. . . . . . n+m
. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .
G oup A – subg oup x VM n+j, n+j+1. . . . . . n+j+m
Simila ly,
G oup B – subg oup 0 VM=0, 1, 2. . . . . . n
G oup B – subg oup 1 VM=n+1, n+2. . . . . . n+m
. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .
G oup B – subg oup x VM n+j, n+j+1. . . . . . n+j+m
and so on.
He e,n+j = alue ID 1 inc emen s han las one in he p e ious subg oup
A lag indica es i a subg oup is ull o no
3.2 G oup Based Resou ce Managemen
The esou ce managemen p ocedu e has di e en po ion ha wo ks oge he o a be e cos
e icien managemen o he se e s. This sys em uses Ene gy e icien managemen o IaaS[15]
cloud which is a in eg a ed app oach o VM mig a ion and econ igu a ion, and PM powe
managemen and Task Scheduling algo i hm in Cloud Compu ing En i onmen Based on Cloud
P icing models and [28] ,as a base o ou algo i hms o o e a be e load balancing and esou ce
u iliza ion based on esou ce in ensi y awa e load balancing . The di e en po ion is esponsible
o di e en managing wo k in esou ce managemen .
• Access Log: The access log is esponsible o ensu ing he au hen ici y o he use
eques .
• Assignmen Manage : The assignmen manage is esponsible o assigning use o a
eese ice node which in his case will be a VM.I ensu es easy iden i ica ion and
se icing o he use a ha VM as his ID’s main ains a sequence acco dance wi h hei
g oup.
• Schedule :The schedule is esponsible o assigning a PM which ep esen s an
ins anceo he esou ces ha is equi ed by he use ask. The PM assignmen o a VM
depends onwhich g oup hey a e assigned in. I p og ams a VM mig a ion o o he g oup
when he e isno enough esou ce in he g oup.
• Load S a is ics: This po ion check o load in each VM and e en ually in each g oup.
This coun s he load as pe cen age o he o al esou ces assigned and he amoun
occupied. This uses a h eshold powe o de e mine i his g oup can ha e any new use s
o new PM could be assigned o a oid any collusion.
In e na ional Jou nal o Compu e Science & In o ma ion Technology (IJCSIT) Vol 10, No 4, Augus 2018.
30
• P edic o : The p edic o gi es an es ima ion on i he e is any possibili y o con lic and
hunge among use s can happen.
• P icing and e enue: Calcula es he p ice o he use .Also shows p ice ha he use
will ind i hey use some new esou ces. Wi h he p o ide ’s inpu in each g oups cos
he e enue po ion calcula es he mon hly e enue om a pa icula g oup which is
calcula ed as pe subg oup.
Figu e 2: G oup based esou ce managemen
3.3.Resou ce G ouping and P o i maximiza ion
The cloud se ice p o ide ’s ac ual p o i e e s o he di e ence be ween he e enue and
hecos s[23].To maximize se ice p o ide s’ p o i s and ensu e load balancing wi hin hem,
eachcloud p o ide de e mines he in-sou cing p ice as well as he a ailable quo a o esou ces
basedon which g oup he se ing esou ces a e esided.The p icing policy p oposed in his pape
uses he insou cing p ice which is se acco ding o he VM cos s o a ce ain ype o
asks.Di e en g oup has di e en p ice calcula ed wi hin a iming in e al.G oups wi h lowe
powe has lowe p ice and highe powe equi ing g oups has p ice se based on he ype o
esou ces assigned o hem.Acco ding o cos based p icing model se ices a e p iced depending
on he esou cesconsumed.The e o e each use ask in his model is p iced based on he g oup
hey a e assigned o.As a esul di e en asks can ge easily p iced based on hei powe
equi emen a he hanha ing a common p ice calcula ed o e ime.The p oposed cons ain s
ensu es ha only limi ednumbe o PMs a e ac i a ed and a he same ime asks a e ha ing low
a e age wai ing ime in he queue.G ouping and subg ouping o esou ces ensu es e enly
dis ibu ion o esou ces among he execu ing asks hence educing unde o o e consump ion o
In e na ional Jou nal o Compu e Science & In o ma ion Technology (IJCSIT) Vol 10, No 4, Augus 2018.
31
compu ing powe .This educes he o e all p o isioning cos and inc easing he g oss p o i o he
se ice p o ide .
4.
G
ROUP
B
ASED
R
ESOURCE
M
ANAGEMENT AND PRICING
This sec ion desc ibes how Resou ce g oup model is applied o di e en aspec s o esou ce
managemen which includes ask assignmen , ask scheduling, load balancing along wi h p icing
o p o isioning esou ces.Di e en algo i hms ha e been designed o each sec ion along wi h
necessa y cons ain s.
4.1. Task Assignmen
This sec ion depic s how asks a e assigned o di e en esou ce g oup. The main idea o he
enhanced algo i hm is alloca ing he pa icula VM o he eques ed asks acco ding o he
ollowing s eps:- A i s he equi ed compu a ion powe o each VM is calcula ed based on he
o al numbe o MIPS alloca ed o i . Calcula e he p ocessing powe o each VM
i
(i.e.,VM
i
’s
To al MIPS) using equa ion (1)
To al_MIPS[i] = CPUs
∗
MIPS (1)
CPUs= he numbe o co es, MIPS=million ins uc ion pe second o he single co e o he VM.
He e i=0, 1, 2. . . . . . , ep esen s he posi ion o he subg oup. Then calcula e he equi edpowe
o each ask o all he ecei ed asks om he use using equa ion (2).I he wai ing askshas m
no o ’p’ ype o asks hen
=
∑
(2)
Di e en h eshold alues a e de e mined o di e en g oups. Tasks a e assigned o he g oup
which ma ches wi h i s powe equi emen .Such as o g oup A i a ask o ype p equi es
MIPSsuch ha MIPS
p
(i)<Th eshold
A
, i ’s assigned o g oup A. The e o e all m asks o ype p
will beassigned o G oup A a e inspec ing which subg oups o choose.
An a ay is used o o e hese e nodes in a g oup. Then using equa ion (2) i is de e mined i
he e is enough esou ces e eni he e is a ee node.A Time h eshold T
lowcon
=Low powe
consump ion o ime e e s i anode is consuming a e y low se e powe (low-con) o a
ce ain ime T hen he VM should be ans e ed o low powe consump ion g oup o a oid
was age and ice e sa o ime T
high_con
=High Powe Consump ion o ime
Acco ding o he calcula ions he ollowing algo i hm is designed o assignmen o he asks
wai ing in he queue o execu ion wi h conside ing he powe equi emen .
Algo i hm 1: Task Assignmen algo i hm
1.
p ocedu e
Assignmen (G; s) //
G g oups, s= asks o be assigned
2. A ay[Se ice_node(ni)] //use assigned iden i ica ion no
3. A ay[a][b]
4. a = use _id
5.
Calcula e MIPS o all p ype o asks
6.
o
j = 0 o o al_numbe o _g oups
do
//
de e mining he equi ed g oup
7. Compa e Powe o Task
p
wi h g oup h eshold
8. o all subg oups in g oup G do
9. i ( henload_in_subg oup>T
sa e_load
)
10. choose g oup G
11. elseWai in he queue
In e na ional Jou nal o Compu e Science & In o ma ion Technology (IJCSIT) Vol 10, No 4, Augus 2018.
32
12. end i
13. end o
14. end o
15. Check o he nex eques
16. end p ocedu e
4.2. Task Scheduling
A e selec ing a g oup he asks a e needed o assign o a VM con ained wi hin a
subg oup.Powe ac o o each VM o a subg oup x is calcula ed as ollows-
PF o VM,VM[i]
PF
=
!"#$%&_"
_"
(3)
Powe ac o o each subg oup is calcula ed using equa ion (4) which de ines he c edibili y o
each subg oup
PFo subg oups ,subg p[n]
PF
=∑
(
!"#$%&_)
_)
&
)*%
(4)
He e,n=no o subg oup,s= i s VM in he n h subg oup and e=las VM in n h subg oup. I PFo a
subg oup is g ea e han x he subg oup is no eligible o new ask scheduling. Ra he
hansea ching each VM sepa a ely o eligibili y his mechanism allows a g oup o VMs o be
iden i ieda a ime which educes he ime cos o sea ching sui able VM.Tasks a e so ed
acco ding o hei p io i y in he wai ing queue.Requi ed powe is calcula ed o he ask wi h
equa ion (2) and i ’spowe ac o is calcula ed using equa ion(5).
Task s
01
=
2345
6
7
89:
(5)
Sea ch he eques ed subg oup sequen ially o ind a VM ha p o ides p ocessing powe equal o
o less han he powe ac o o he ask by conside ing ha he di e ence be ween he
selec ed asks p ocessing powe and VM o be minimum.
Algo i hm 2: Task Scheduling Algo i hm
1:p ocedu e Schedule (n, s)
2: o i = 0 o m do
3:De ine MIPS o he VMs
4:Calcula e he powe ac o o each VM
5:end o
6: o i = 0 o ndo
7:De ine MIPS o he subg oups
8:Calcula e powe ac o o each subg oup
9:end o
10: o i = 0 o m do
11:Calcula e he PF o each ask
12: Choose he VM wi h which ask ha e he highes powe ac o VM
i
13:allo men =Task
PF
* eques ed MIPS
14:end o
15:end p ocedu e
In e na ional Jou nal o Compu e Science & In o ma ion Technology (IJCSIT) Vol 10, No 4, Augus 2018.
33
4.3. Load Balancing
O e loaded PM ans e s VMs unning on hem which ha e high consump ion on high-powe
esou ces and limi ed consump ion on low powe esou ces.The e o e i elie es i s load quickly
a he same ime comple ely u ilizing da acen e esou ces.Selec ed PM gene ally has high
compe ence on he high in ensi y esou ces o ac i ely end o igno e o e loading des ina ion
PMs in he u u e.In ou wo k ask alloca ion and VM alloca ion wo ks based on communica ion
a e. Communica ion a e e e s o he numbe o con ac s be ween he sou ce VM and des ina ion
VM in a uni ime pe iod.Communica ion a e T
xi
(VM
i
o subg oup x) o a VM
ni
wi h i s local
VM
kh
(VM
h
o anysubg oup k) is deno ed by,
;"
∑
;"<=
#
"*>
(6)
k=x i he wo communica ing VM a e in he same g oup.Task e-alloca ion akes place wi hin he
same subg oup whe eas VM e-alloca ion occu s among di e en g oups.When VMs and asksa e
ealloca ed he pe o mance deg ades[27]. We aim o minimize he deg ada ion.To educe
hedeg ada ion he numbe o swi ches in he communica ion pa h need o be
lessened.Realloca iondepends on he esul s o equa ion (6) and (7) as ollows Load o VM
i
compa ing wi h o he s in he same subg oup
?
@
=
@"
A
%$BCD#
A
(7)
I L
VMi
< h eshold
load
(minimum load: below his h eshold is conside ed as low powe consuming)
hen, asks o VM
i
is ansmi ed o VM
j
whe e
?
@
E
= min(
@J
A
@"
A
) (8)
i VM
j
has enough esou ces and has low communica ion a e wi h he ansmi ing VM
o he wisei swi ches o nex VM.The e o e he now ee VM and i s ela ed PMs u ned o s and
by mode.Simila ly i a subg oup is unning low consuming VMs hey a e ans e ed o he
subg oup wi hminimum communica ion cos .In ou p oposed me hod he numbe o swi ches is
cons ained by henumbe o VMs in a subg oup o asks ealloca ion and numbe o subg oups
o VM alloca ion. I one node is consuming mo e powe i ’s ask is dis ibu ed. To dis ibu e he
ask load ee esou cesa e a i s sea ched wi hin he g oup i belongs o. I ensu es esou ce
u iliza ion and keeps mo ePM a s andby mode. I a node A is consuming oo much CPU powe
and he consecu i esubg oup is al eady c ossing i s h eshold hen o u he assignmen he VM
is ans e ed o henex subg oup ha can ake i .
Algo i hm 3: Load balancing Algo i hm
1: p ocedu eload_balancing
2: o i = 0 o ndo
3: Check load in each subg oup
4: i ( henLoadinSubg oup[n]>T
accep ed
)
5: whileLoadinSubg oup[n]>T
accep ed
do
6: Selec he maximum powe consume is ,k
7: o each node in he subg oup do
8: Calcula e he communica ion a e wi h k
9:
Choose he VM
wi h wi h highes communica ion a e and low load
10:
Shi ask o he chosen VM
11: end o
12: end while
13: end i
14: i LoadinSubg oup[n]<T
accep ed_low
hen
