Performance Optimization of Microservice Applications under Resource Constrained Environments

Performance of microservice applications deployed on cloud platforms have a non-linear relationship with resources allocated to each service of the application. Applications can be limited by fixed resource budgets or operation costs. This study presents an automated framework utilizing equality constrained Bayesian Optimization (BO) on CPU, and Memory limits of individual services in a benchmark single node microservice application with the objective of minimizing latency and maximizing throughput. The model found configurations that achieve over 3 times improvement on latency and over 2 times improvement on the throughput of the default configuration. Particle Swarm Optimization (PSO) achieved similar improvement in performance with a higher number of iterations compared to BO.