Optimizing Microservice Design Parameters

Abstract

Production-level cloud services are increasingly deployed as microservices. An important question is given application logic, how to design an effective microservice architecture. Existing studies have underscored the importance of microservice cohesiveness and coupling, using these metrics to drive automatic design optimizations. However, they have not accounted for the potential impact that such design changes may have on overall system performance, which is confirmed by our case study. In this work, we present a system that can automatically identify microservice designs that are well-balanced across performance, coupling, and cohesiveness to meet cloud provider’s requirements. the system uses a multi-round dynamic programming approach, selectively identifies promising design candidates, generates the corresponding microservice code, measures and compares the results to ultimately determine the optimal design. The designs produced by our system typically achieve over 20% throughput improvement under the same QoS with less than a 10% increase in average LCOM, and often outperform the original benchmark architectures across all evaluated metrics.