Cost minimization for deploying serverless functions

Abstract

The costs of serverless functions increase proportional to the amount of memory reserved on the deployed server. However, increasing the amount of memory decreases the function execution time, which is also a factor that contributes to cost. We propose an automated approach for optimizing the amount of memory reserved for serverless functions. First, we measure the running time of a given function in various memory settings and derive a regression model. Then, we define an objective function and a set of constraints based on this regression model and the configuration space. Finally, we determine the optimal memory setting for minimizing cost. Our industrial case study shows that significant cost reductions can be achieved by accurate estimations of the impact of memory settings on runtime performance.