Energy and QoS-Aware Microservice Resource Management Using Multi-Objective Multi-Task Reinforcement Learning

Abstract

The paper talks about microservice architecture. This kind of design splits a big application into small services. Each service works on its own. These services are able to be placed on different servers. This makes systems more flexible and scalable. Most past works only focus on reducing delay. But delay is not the only concern. Power use is important. High power use leads to high cost and poor sustainability. So, this paper looks at both delay and power together. A new system termed GIRP was proposed. GIRP stands for Green and Intelligent Resource Provisioning. It tries to manage both delay and power. It does this by controlling CPU usage and service copies. These are placed on servers smartly. To make good decisions, GIRP uses a special learning method. This method is titled as Deep Reinforcement Learning. It learns by trying different actions. It sees what works best. Then it learns from that.  In this work, a multi-Task learning model is used. It decides two things at the same time: the number of CPU cores to use and the number of service replicas to create. This helps in managing resources well. Another part of the system handles placement for each service. GIRP is tested using a popular benchmark. The results are strong. This proves GIRP works well. This paper shows a smart way to manage microservices. It saves power and meets user delay needs. The learning model makes better choices over time.