Restructuring microservice systems with ordered messaging and parallel process flow approaches for performance optimization

Abstract

Microservice systems are composed of small services with a single responsibility, which can be independently developed and scaled. The design has the disadvantage of increased system complexity because of the introduced remote communication for service interaction. Communication patterns strongly influence microservices system architecture and performance. They strengthen or weaken the independence of services, contributing to system availability and reliability. The system's load increases with the introduction of new users, but the system's performance still needs to meet agreed service-level objectives.

The thesis aims to analyze the impact of changes in communication patterns and technologies on the microservice system performance on the example of the Connected Retail Inbound pipeline. Performance is defined in terms of latency and throughput of the system fragment.

The study describes two approaches for enhancing the performance of microservice systems by restructuring: ordered messaging and parallel process flow. The work is limited to trialing the approaches on only one microservice system. Theoretical sources suggest that both techniques would improve the performance. However, the implementation of the ordered messaging method had an adverse effect on the pipeline message processing performance in terms of throughput. The parallel process flow approach improved the pipeline performance in terms of the average worst message processing latency. Queueing theory models are used to evaluate observed results.