Cluster resource management and task scheduling method and system based on deep reinforcement learning

Abstract

The invention relates to a cluster resource management and task scheduling method and system based on deep reinforcement learning. The method comprises the steps: placing tasks needing to be operatedin a to-be-scheduled task queue; sequentially processing the tasks in the to-be-scheduled task queue through a resource scheduling management intelligent agent, and generating a scheduling decision according to cluster resource conditions and resource requirements of the tasks, wherein the resource scheduling management intelligent agent is a neural network obtained by training by using a deep reinforcement learning method according to historical task records running on a cluster; and according to the scheduling decision, scheduling the task to a corresponding machine in the cluster for execution. According to the invention, the utilization rate of cluster resources and the throughput rate of the system can be improved, and computer cluster resource allocation can be self-adaptive when thetask load condition changes; according to the method, the task response time can be shorter, the number of cluster machines can be reduced under the condition of the same load, and the method and system have important significance in saving energy and protecting the environment.

Description

technical field [0001] The invention belongs to the technical field of computer software, and relates to a resource management and task scheduling method and system for large-scale computer clusters. Background technique [0002] In today's information age, people cannot live without the various services provided by the Internet, which has also spawned a large number of giant Internet companies, such as Alibaba, Baidu, Tencent, Meituan, etc.; at the same time, cloud computing and The concept of the Industrial Internet of Things is also booming. Originally scattered small computer clusters are gradually migrating to large clusters of cloud computing vendors, such as Alibaba Cloud and AWS. This makes large Internet companies and cloud computing vendors need to maintain a large number of computer clusters and countless applications and tasks running on them. Maintaining the normal operation of a large number of clusters requires a lot of power consumption, which is an importan...

AI Technical Summary

Problems solved by technology

[0004] Since task scheduling and resource management are essentially a bin packing problem, this problem has been proven to be an NPC problem and cannot be optimally solved in polynomial time

The main problem of the existing methods based on heuristic and artificial rules is: on the one hand, heuristic rules are easy to fall into the local optimal solution, and cannot achieve a good box packing effect, that is, there are many fragmented resources, and the utilization of system resources The rate is not high; on the other hand, when the task composition or cluster structure changes, the method based on expert compilation of rules needs to be explored again, which wastes a lot of manpower and material resources.

What is more difficult is that, different from the simple packing problem, the task scheduling and resource management of computer clusters usually have a variety of goals, such as shortening task waiting time, improving resource utilization, and providing data locality support. Some goals are mutually influencing and restricting each other, and existing methods cannot take into account multiple goals at the same time

Images