Microservice architecture enhances scalability and flexibility by decomposing a single application into loosely coupled, independently deployable components. Unlike monolithic services, microservices require frequent inter-component communication and are deployed with multiple replicas primarily to ensure quick response times to varying user requests. However, existing studies have not adequately addressed the challenges inherent in multi-replica environments, leading to suboptimal performance. To address these challenges, we propose a latency-aware replica placement method and traffic allocation strategy that minimizes communication overhead among replicas while achieving load balancing. These issues are especially critical in edge-cloud environments with diverse and high inter-node latencies, leading to QoS degradation. The experimental results show significant reductions of up to 81% in the 99th percentile tail latency compared to existing techniques.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Optimizing Traffic Allocation for Multi-replica Microservice Deployments in Edge Cloud

  • Hokun Park,
  • Hyungjun Kim,
  • Donggyun Kim,
  • Gyujeong Lim,
  • Heonchang Yu

摘要

Microservice architecture enhances scalability and flexibility by decomposing a single application into loosely coupled, independently deployable components. Unlike monolithic services, microservices require frequent inter-component communication and are deployed with multiple replicas primarily to ensure quick response times to varying user requests. However, existing studies have not adequately addressed the challenges inherent in multi-replica environments, leading to suboptimal performance. To address these challenges, we propose a latency-aware replica placement method and traffic allocation strategy that minimizes communication overhead among replicas while achieving load balancing. These issues are especially critical in edge-cloud environments with diverse and high inter-node latencies, leading to QoS degradation. The experimental results show significant reductions of up to 81% in the 99th percentile tail latency compared to existing techniques.