Intelligent task scheduling strategy training method based on strategy gradient reinforcement learning

Abstract

The invention relates to an intelligent task scheduling strategy training method based on strategy gradient reinforcement learning. The method mainly comprises task scheduling sequence data generationbased on reinforcement learning and a task scheduling strategy training method based on strategy gradient reinforcement learning. According to the method, a task scheduling problem is abstracted intoa reinforcement learning process, a strategy gradient descent training network model is used, a scheduling strategy training method combining a multi-dimensional task state matrix, a scheduling strategy matrix and a reward function is provided, the training efficiency of a task scheduling strategy is improved, and the total running time of the task is shortened.

Description

technical field [0001] The invention relates to the technical field of computer system task scheduling and artificial intelligence algorithms, in particular to an intelligent task scheduling policy training method based on policy gradient reinforcement learning. Background technique [0002] Facing the big data analysis and computing tasks in the data center, how to efficiently schedule computing tasks to run on large-scale servers has become a key issue in computer systems. Big data analysis and computing tasks have the characteristics of high parallelism and complex data dependencies, and the task scheduling problem is facing huge challenges. At the same time, there are large performance differences among server computing devices in data centers, thus forming a complex distributed heterogeneous computing system. Scheduling parallel computing tasks in a distributed heterogeneous computing system is a recognized NP-hard problem in the industry, and it is impossible to find ...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the above-mentioned deficiencies, the present invention aims to provide a strategy training method for intelligent task scheduling based on strategy gradient reinforcement learning. This method abstracts the task scheduling problem into a process of reinforcement learning, and uses strategy gradient descent to train the network model. A scheduling policy training method combining multi-dimensional task state matrix, scheduling policy matrix and reward function is proposed, which improves the training efficiency of task scheduling policy, reduces the overall running time of tasks, reduces the consumption of computing resources, and solves the problems in the prior art. Existing task scheduling policy training complexity is high, the scheduling policy is sub-optimal, and consumes a lot of computing resources.

Images