Abstract

Serverless has emerged as the next dominant cloud architecture and programming paradigm due to its high scalability and flexible, pay-as-you-go billing model. In serverless, developers structure their cloud services as a set of functions connected in a workflow, whereas providers take responsibility for scaling each function’s resources according to traffic changes. This labor division opens great opportunities for systems researchers who seek to innovate in serverless computing. Unfortunately, leading serverless providers, like AWS Lambda and Azure Functions, rely on proprietary infrastructure. While existing in other computer systems domains, the gap between production systems and research prototypes is particularly large for serverless computing, preventing academics from discovering and solving relevant problems in realistic scenarios.

I introduce vHive, an open-source ecosystem for serverless benchmarking and experimentation, with the goal of enabling researchers to study and innovate across the entire serverless stack. vHive includes a holistic performance analysis methodology and a framework that integrates industry-leading components, such as Kubernetes orchestrator and AWS Firecracker hypervisor technologies. To demonstrate vHive’s utility, I analyze the bottlenecks, namely function cold-start and cross-function data communication delays, in the leading commercial clouds. I then present two novel approaches that address each of these problems. First, to reduce the cold-start delays, I find that serverless functions operate over stable working sets – across function invocations – that can be effectively recorded and prefetched. Second, to accelerate data movement, I introduce a serverless-native cross-function data communication fabric that combines high performance with autoscaling capabilities.