Abstract

In recent years, internet enterprises have transitioned from traditional monolithic service to microservice architecture to better meet evolving business requirements. However, it also brings great challenges to the resource management of service providers. Existing research has not fully considered the request characteristics of internet application scenarios. Some studies apply traditional task scheduling models and strategies to microservice scheduling scenarios, while others optimize microservice deployment and request routing separately. In this paper, we propose a microservice instance deployment algorithm based on genetic and local search, and a request routing algorithm based on probabilistic forwarding. The service graph with complex dependencies is decomposed into multiple service chains, and the open Jackson queuing network is applied to analyze the performance of the microservice system. Data evaluation results demonstrate that our scheme significantly outperforms the benchmark strategy. Our algorithm has reduced the average response latency by 37%-67% and enhanced request success rate by 8%-115% compared to other baseline algorithms.

中文翻译：

---

云数据中心微服务部署与路由联合设计

摘要

近年来，互联网企业从传统的单体服务转向微服务架构，以更好地满足不断变化的业务需求。然而，这也给服务提供商的资源管理带来了巨大的挑战。现有研究尚未充分考虑互联网应用场景的请求特征。一些研究将传统的任务调度模型和策略应用于微服务调度场景，而另一些研究则分别对微服务部署和请求路由进行优化。在本文中，我们提出了一种基于遗传和局部搜索的微服务实例部署算法，以及一种基于概率转发的请求路由算法。将具有复杂依赖关系的服务图分解为多个服务链，并应用开放的Jackson队列网络来分析微服务系统的性能。数据评估结果表明，我们的方案明显优于基准策略。与其他基线算法相比，我们的算法平均响应延迟降低了37%-67%，请求成功率提高了8%-115%。