Benchmarking microservice performance: A pattern-based approach

Abstract

Benchmarking microservices serves to understand and check their non-functional properties for relevant workloads and over time. Performing benchmarks, however, can be costly: each microservice requires the design and implementation of a benchmark, possibly repeatedly as the service evolves. As microservice APIs differ, benchmarking tools that assume common interfaces - like ones for databases - do not exist. In this work, we present a pattern-based approach to reduce the efforts for defining microservice benchmarks, while still allowing to measure qualities of complex interactions. It assumes that microservices expose a REST API, described in a machine-understandable way, and allows developers to model interaction patterns from abstract operations that can be mapped to that API. Possible data-dependencies between operations are resolved at runtime. We implement a prototype of our approach, which we use to demonstrate that it can be applied to open-source microservices with little effort. Our work shows that pattern-based benchmarking of microservices is feasible and opens up opportunities for microservice providers and tooling developers.