Home
Serverless computing has seen rapid adoption because of its instant scalability, flexible billing model, and economies of scale. In serverless, developers structure their applications as a collection of functions, sporadically invoked by various events like clicks. The high variability in function image sizes, invocation inter-arrival times,and request burst sizes motivates vendors to scrutinize their infrastructure to ensure a seamless user experience across all of their services.
To monitor serverless platform performance, identify pitfalls, and compare different providers, the public attention has turned to benchmarking, whereby measurement functions are deployed to the cloud to gather insights regarding response latencies, transfer speeds, and vendor policies. This work introduces our open-source framework for serverless performance evaluation, intending to enable researchers and developers to benchmark multiple cloud platforms. Using this framework, we conduct one of the biggest serverless measurements to date, launching over 320,000 function instances to characterize system performance in AWS Lambda and vHive - the Edinburgh Architecture and Systems (EASE) virtual machine orchestrator system.