Cooling Down FaaS: Towards Getting Rid of Warm Starts

Abstract: Serverless execution and most notably the Function as a Service (FaaS) model got quite some attention during the recent years. As of today, all commercial and open source implementations follow the common practice of keeping the execution environments running to achieve low function execution latency. In this paper we compare the startup latency of different available virtualization technologies, then we implement and benchmark an FaaS prototype system using IncludeOS unikernels for function execution. We show that our system can start and execute functions with essentially the same latency as AWS Lambda with its continuously running executor units. Due to the low overhead, this approach opens the possibility for simplified FaaS platforms without the resource waste and extensive monitoring requirements of existing solutions.