Limitations of AWS and GCP Serverless Functions

Abstract

Cloud providers offer serverless functions such that users can focus on the business logic and worry only about the code and application, leaving scaling, security, software updates, and everything else to the cloud provider. However, the minimal user interactions with the configuration of the functions may result in errors executing the code or may not scale properly. In this research, we analyze different implementations of a complex cloud function that requires execution in less than 10 seconds and analyze the responses from AWS and GCP serverless platforms. We observed that AWS can activate a maximum of 1000 instances, although they claim a higher number, while GCP can reach a maximum of 3000 active cloud instances. However, GCP faces another problem of activating only a smaller number of cloud instances for executing small-sized functions. This limitation might surprise the user since it generates a lot of errors with a “request timeout” response. We elaborate on lessons learned by comparing AWS and GCP functions with details on the bad performance of GCP Cloud Functions in our experiments.