Seminar:
Intelligent Serverless Computing Systems

ABSTRACT

Serverless computing has proliferated as a new cloud computing paradigm that offers ease-of-management, cost-efficiency, and auto-scaling by packaging executable codes as lightweight functions. Numerous applications have been shipped as serverless functions, including data analytics, video processing, scientific computing, HPC, and machine learning. Unfortunately, existing serverless computing platforms require users to pre-define resource allocations for their functions, leading to frequent misconfiguration by inexperienced users in practice. Besides, functions’ varying input data further escalate the gap between their dynamic resource demands and static allocations, leaving functions either over-provisioned or under-provisioned. In this talk, I will first briefly reveal the observations of idle resources in serverless computing. Then, I will talk about Freyr (accepted to WWW 22), a novel resource manager driven by deep reinforcement learning that accelerates function executions by harvesting idle resources. Next, I will introduce Libra (accepted to HPDC 23), an input data size-driven resource harvesting framework that harvests idle resources safely and timely for serverless computing. Lastly, I will share our recent works and vision of improving serverless computing.

Hao Wang

Louisiana State University

Dr. Hao Wang is an Assistant Professor in the CSE Division at the Louisiana State University. He received both of his B.E. degree and M.E. degree from Shanghai Jiao Tong University, China and the Ph.D. degree from the University of Toronto, Canada. He was the receipt of NSF CRII Award in 2022. Dr. Wang leads the IntelliSys Lab at LSU, and his research focuses include serverless computing and federated learning.