PhD Proposal by Adam Hall
Title: Lifecycle Management of Execution Environments for the Edge
School of Computer Science
College of Computing
Georgia Institute of Technology
Date: Friday, July 22, 2022
Time: 11:00 AM - 12:30 PM (EST)
Dr. Umakishore Ramachandran (Advisor, School of Computer Science, Georgia Institute of Technology)
Dr. Alexandros Daglis (School of Computer Science, Georgia Institute of Technology)
Dr. Alexey Tumanov (School of Computer Science, Georgia Institute of Technology)
Dr. Tushar Krishna (School of Electrical and Computer Engineering, Georgia Institute of Technology)
Dr. Sameh Elnikety (Senior Principal Researcher, Microsoft Research)
Edge computing, the notion of placing computational resources at the last-mile of the network (such as within a wiring closet or on a cell phone tower), has emerged as a way to serve the needs of next-generation applications requiring high throughout and low latency. Unlike the Cloud, resources within Edge computing nodes are necessarily limited due to physical space constraints. Because of this limitation, traditional isolation methods for hosting applications on the same platform, such as dedicated virtual machines, are unsuitable due to the resource overhead they require. Alternative methods, such as process-based containers and language-based sandboxes, offer logical application isolation with much less overhead but suffer from issues such as long startup delays and slower-than-native execution speeds. Additionally, these methods lack adequate performance isolation mechanisms that enable efficient resource sharing while ensuring consistent low latency application response times. To achieve a high degree of multi-tenancy, the Edge needs a lightweight execution environment that segments application instances and imposes resource restrictions with minimal overhead.
We propose the development of a hybrid runtime for the Edge that leverages the strengths multiple execution environments and incorporates performance isolation techniques to increase the efficiency of application components executing on Edge platforms. Specifically, our proposed runtime will demonstrate by proof-of-construction the feasibility of a hybrid approach in lowering average and tail response time latencies of applications and increasing resource utilization of the platform on which they execute. We will evaluate our solution using a combination of peer-reviewed benchmarks targeting serverless and microservice architectures coupled with real-world trace data from these services.