A multi-resource cloud job scheduling method based on deep Q-network algorithm

Abstract

The present invention provides a multi-resource cloud job scheduling method based on the Deep Q-network algorithm, which includes four steps: collecting current configuration information of resources and job requirement information through the cloud environment; current configuration information of resources and job requirements The information is represented by a matrix image, which includes cells, cells of the same color represent the same job, and the rectangle formed by cells of the same color includes M×N cells, M represents the number of resources, and N represents the time step ; According to the matrix image, using a deep learning method to obtain high-level semantic information; according to the high-level semantic information, using a reinforcement learning method to complete real-time scheduling and planning of resources. The matrix image of the present invention can fully and clearly represent the status of resources and operations. This method also uses deep reinforcement learning, which combines depth and reinforcement learning, in which deep learning mainly completes environmental state perception, while reinforcement learning completes decision-making and realizes the mapping from state to action.

Description

Technical field [0001] The invention relates to the field of cloud computing resource scheduling, in particular to a multi-resource cloud job scheduling method based on the Deep Q-network algorithm. Background technique [0002] Resource scheduling is a difficult and hot research topic in the field of cloud computing. A good resource allocation and scheduling strategy can effectively utilize resources and increase the economic benefits of suppliers while ensuring user quality of service (QoS). Cloud computing resource scheduling is actually a multi-constrained, multi-objective optimization NP-hard problem. At present, the traditional method of solving decision-making problems is to design efficient heuristic algorithms with performance guarantees under specific conditions. The generality and practicability are not strong, and they cannot adapt to the changing and complex cloud environment. In addition, the researchers abstracted resource scheduling in the cloud environment into...

AI Technical Summary

Problems solved by technology

The disadvantage of this patent is that the active virtual machine data cannot fully represent the status of resources and jobs in the cloud environment

Images