Communication Affinity Aware Multi-Constrained Container Colocation

Abstract

Abstract Microservice-based systems are increasingly adopted due to their scalability and flexibility. However, analyzing their communication patterns and optimizing container network placements present significant challenges. This thesis addresses these challenges by proposing a comprehensive methodology and tools for empirical analysis and optimization of microservice communication patterns. To facilitate this analysis, we developed MockNet, a microservice framework designed for easy monitoring, logging, and replaying of microservice communication patterns. However, analyzing the extensive logs generated by MockNet manually became impractical as the system scaled. To address this issue, we introduced the Bridge data processing pipeline, enabling automated evaluation of various scheduling algorithms, including our proposed Multi-Constrained Container Colocation (M3C) scheduling algorithm. Using the standardized Crewmen workflow, we were able to quantify container network optimizations consistently across di↵erent microservice implementations and scheduling algorithms. Our analysis revealed that the M3C scheduling algorithm outperforms existing state-of-the-art algorithms, achieving higher container network optimization with fewer colocations, thus reducing the service downtime. Additionally, we implemented the Crewmen container orchestrator, which supports the M3C scheduling algorithm. However, it is worth to notice that Crewmen is not yet recommended for the production use, since real-world container orchestration considerations such as security, Quality of Service (QoS), and scalability are yet to be addressed. In conclusion, this research provides a standardized methodology for analyzing and optimizing container networks in microservice-based systems. The introduced definitions, tools and algorithms, including Crewmen Workflow, MockNet, Bridge, M3C scheduling algorithm, and Crewmen orchestrator, contribute to a deeper understanding and optimizations of container networks.