The Role of Public Cloud on AI and Edge Research: A CloudBank Perspective

The Role of Public Cloud on AI and Edge Research: A CloudBank Perspective 1st Shava Smallen San Diego Supercomputer Center University of California, San Diego La Jolla, USA [email protected] 2nd Nicole Wolter San Diego Supercomputer Center University of California, San Diego La Jolla, USA [email protected] 3rd Rob Fatland eScience Institute University of Washington Seattle, Washington [email protected] Abstract—CloudBank is a cloud access entity that supports 324 research awards on public cloud platforms. The majority of CloudBank’s awards involve artificial intelligence and/or edge computing. Here we provide remarks on the nature of public cloud uptake in the computer science research community as it relates to AI and edge computing. Index Terms—public, cloud, research, computing, edge, artificial, intelligence, cost, optimization I. INTRODUCTION CloudBank [1], [2] is a cloud access entity funded by the National Science Foundation (NSF) to help researchers access public cloud resources and has been supporting Amazon Web Services [3], Google Cloud [4], Microsoft Azure [5], and IBM Cloud since August 2020. The main benefits to researchers are that CloudBank enables them to easily access multiple public clouds, share access with collaborators, monitor spend, and access researcher specific training materials and front-line help desk support. Cloud funds flow directly to CloudBank from the NSF with no indirect costs and billing is managed completely by CloudBank’s financial operations so researchers never have to worry about paying monthly bills. Researchers mainly access CloudBank using two methods: v1) NSF researchers can access CloudBank by submitting a proposal to a CloudBank-eligible solicitation and setting aside cloud funds from their proposal budget or v2) NSF researchers that have active awards funded by the Computer and Information Science and Engineering (CISE) directorate can submit cloud requests (up to $100K) directly to CloudBank using a website form (via DCL 22-087). Today, CloudBank averages $230K in spend across 320 NSF awards where 91% (290 awards) are awarded under the v2 model. Of those 290 v2 awards, 73% (212 awards) are related to artificial intelligence (AI) and 9% (25 awards) are related to edge computing with 6% (16 awards) related to both. This paper analyzes the cloud usage patterns awards of AI and edge computing related NSF awards managed by CloudBank under the v2 model. II. CLOUD USAGE PATTERNS When a researcher submits a request to CloudBank, they must estimate their costs for one year using one of the public cloud cost estimate calculators. Figure 1 shows the distribution of request sizes for AI related awards, edge computing related awards, and awards related to both. As expected, AI awards have the most large and medium awards since they typically employ cloud services that require GPUs, which are more costly than CPUs. Fig. 1. Shows the amounts of cloud funds requested by AI related NSF awards, edge computing related NSF awards, and awards related to both. Figure 2 shows the distribution of cloud accounts by the type of NSF award. Amazon Web Services is used the most (81%) followed by Google Cloud (12%) and then Microsoft Azure (7%). Fig. 2. Shows the types of cloud accounts requested by AI related NSF awards, edge computing related NSF awards, and awards related to both. Fig. 3. Shows the top cloud services used by AI related NSF awards on CloudBank where top services are ranked by number of cloud accounts that use them and spend. Figure 3 shows the top cloud services used by AI related awards over the past year (August 2022 to August 2023), where each award can support multiple cloud accounts. The most highly used services are the virtual machine and storage related services. Amazon Elastic Compute Cloud (EC2) is the highest utilized service accounting for $736K in spend across 94 cloud accounts. The same was true for the 39 cloud accounts used in edge computing related awards. AIspecific cloud services, Amazon Sagemaker [6] and Azure Cognitive Services (or OpenAI tools [7]), are utilized in 20 cloud accounts; OpenAI accounted for $24K in spend and is expected to grow in the coming year with the rapid advancements in large language models [8]. Eleven accounts utilized serverless services in Amazon (Lambda) and Google (App Engine) for pieces of their workflow accounting for $7K in spend. Finally, Figure 4 shows the GPU types used on Amazon Web Services by AI related awards with NVIDIA v100 and T4 being the highest utilized GPUs. Fig. 4. Shows the types of GPUs requested by AI related NSF awards on Amazon Web Services. SUMMARY AI and edge computing related research accounts for the majority of NSF awards supported by CloudBank. Most of the AI related awards (88%) are supported on less than $15K (small) or $60K (medium) funds per year. Amazon Web Services is the most highly used public cloud followed by Google Cloud. While basic virtual machine and storage services account for the majority of cloud services used by these awards, AI-specific cloud services like OpenAI are growing in use and demonstrate the role of public cloud services in academic research. ACKNOWLEDGMENT This work is supported by NSF under award #1925001. REFERENCES [1] M. Norman, V. Kellen, S. Smallen, B. DeMeulle, S. Strande, E. Lazowska, N. Alterman, R. Fatland, S. Stone, A. Tan, K. Yelick, E. Van Dusen, and J. Mitchell, “CloudBank: Managed Services to Simplify Cloud Access for Computer Science Research and Education“. In Practice and Experience in Advanced Research Computing (PEARC ’21). Association for Computing Machinery, New York, NY, USA, Article 45, 1–4, 2021. https://doi.org/10.1145/3437359.3465586. [2] CloudBank Website. Retrieved August 24, 2023 from https://www.cloudbank.org/. [3] Amazon Web Services Home Page. Retrieved August 24, 2023 from https://aws.amazon.com/. [4] Google Cloud Home Page. Retrieved August 24, 2023 from https://cloud.google.com/. [5] Microsoft Azure Home Page. Retrieved August 24, 2023 from https://azure.microsoft.com/. [6] Amazon SageMaker Documentation. Retrieved August 24, 2023 from https://docs.aws.amazon.com/sagemaker/index.html. [7] Azure OpenAI Service Documentation. Retrieved August 24, 2023 from https://azure.microsoft.com/en-us/products/ai-services/openai-service-b. [8] The Rise and Rise of A.I. Large Language Models (LLMs) their associated bots like ChatGPT. Retrieved August 24, 2023 from https://informationisbeautiful.net/visualizations/the-rise-of-generativeai-large-language-models-llms-like-chatgpt/.