Reliable performance metrics for microservice - based application

Abstract

Microservice is the most popular technology nowadays. But unfortunately, performance testing with microservice-based applications was comparatively more challenging than testing. As a result, many studies have focused on the performance testing bottleneck in the Microservice-based application. Furthermore, performance attributes were challenging to measure due to the unique features inherited from microservices, such as auto load balancing auto-provisioning elements.

The research aims to identify the best way to reduce the variation in performance test execution and propose more stable performance metrics that stand more reliable in microservice-based applications. Performance tests execution uses different load patterns to identify the best way to reduce variations in a different implementation. Stress, endurance, and load tests were used primarily as the load patterns.

Key metrics were identified beforehand and evaluated to determine the most reliable metrics. The thesis refers to the suggested future works of the existing research. In addition, refer the same analysis to the detailed studies on the performance characteristics of microservice applications. The research's scope is to identify the possibility of measuring the microservice-based applications' performance and finding the best metrics with each load pattern.

The Benchmark application was TeaStore for the proposed evaluations. Further, the Sock Shop application illustrates the performance aspects comparatively against the TeaStore applications. The proposed evaluation method used the TeaStore and Sock shop applications deployed on Google Kubernetes Engine and JMeter Scripts.

The analysis evaluates metrics collected, graphs, and other statistics in JMeter and Google Kubernetes dashboard.