Exploring Imbalances among Microservice Containers in Large Cloud Platforms.

As software systems evolve, monolithic applications are often transformed into a collection of lightweight and loosely-coupled microservices. Understanding the intricacies of microservices is crucial for optimizing their deployment. Unfortunately, existing research and literature often fail to provide a comprehensive analysis of microservice performance and resource utilization, particularly when it comes to addressing performance imbalances across microservice containers within large-scale production clusters.This paper aims to address a critical research gap by conducting an extensive analysis of performance imbalances within Alibaba’s clusters, utilizing response time as a key indicator. Additionally, we investigate the issue of utilization imbalance. Our study reveals that a substantial number of microservices suffer from performance imbalances, primarily attributed to disparities in node resource utilization and uneven communication overhead resulting from inter-node requests. Furthermore, we identify that offline jobs significantly contribute to the imbalanced utilization of node resources. Based on our findings, we propose a novel joint optimization designed to minimize inter-node requests. By adopting this optimization, we aim to enhance the end-to-end performance of the system.