Please note: This master’s research paper presentation will be given online.
Shahzaib Ali, Master’s candidate
David R. Cheriton School of Computer Science
Supervisor: Professor Tim Brecht
With serverless computing platforms gaining in popularity, cloud providers constantly strive to make efficient use of the hardware infrastructure to provide scalable services that seamlessly handle workloads from multiple users with minimum overheads. One overhead of concern is cold start times which refers to the overheads incurred to load the software required to run a function if it is not yet loaded into memory.
In this project, we describe a discrete-event simulator, PerFaaS, designed to mimic production serverless computing environments. The simulator enables the implementation of different resource scheduling and eviction algorithms that can be used to reduce cold start costs. PerFaaS also tracks a variety of statistics to allow algorithms to use these statistics to make informed scheduling and eviction decisions. The output from the simulator can be used to evaluate and compare these policies and algorithms.