Tuning Runtimes in Open Source FaaS

A dynamically expanding area of cloud computing is Function as a Service (FaaS). FaaS allows customers to develop, run, and manage application functionalities on cloud infrastructure without the burden of building and managing a virtual infrastructure. The vast majority of FaaS services used in production are provided by public cloud operators, but a growing number of open source FaaS frameworks offer an alternative deploying on-premises FaaS services. FaaS frameworks support different programming language runtimes. The performance of such systems is dependent on these language runtimes. Our goal is to show and analyze this dependency, and provide insights to the important aspect when performance is essential. This paper provides a measurement based evaluation of the capabilities of different language runtimes in FaaS frameworks. We evaluate three different workloads (echo, compute intensive, and data intensive) on the selected runtimes.