FaaStest - Machine learning based cost and performance FaaS optimization

Abstract

With the emergence of Function-as-a-Service (FaaS) in the cloud, pay-per-use pricing models became available along with the traditional fixed price model for VMs and increased the complexity of selecting the optimal platform for a given service. We present FaaStest - an autonomous solution for cost and performance optimization of FaaS services by taking a hybrid approach - learning the behavioral patterns of the service and dynamically selecting the optimal platform. Moreover, we combine a prediction based solution for reducing cold starts of FaaS services. Experiments present a reduction of over 50% in cost and over 90% in response time for FaaS calls.