Multi-target cloud workflow scheduling method based on joint reinforcement learning strategy

Abstract

The invention discloses a multi-target cloud workflow scheduling method based on a joint reinforcement learning strategy, and the method comprises the steps: building a reinforcement learning agent joint strategy model through the extension of the attributes and methods of a workflow request and a cloud resource, and enabling the scheduling model to be more suitable for an actual workflow application scene. According to the method, the influence of the scheduling process, each decision sub-network and historical decision information is comprehensively considered during behavior selection, so that the finally selected behavior is more reasonable, the dominance and diversity of generating a non-dominated solution set by the algorithm are further improved, and the practicability of the method is effectively improved.

Description

technical field [0001] The invention belongs to the technical field of cloud computing, and in particular relates to a multi-objective cloud workflow scheduling method based on a reinforcement learning strategy. Background technique [0002] As a new paradigm of distributed system computing, the pay-per-use and elastic resource model of cloud computing provides an easily accessible and scalable infrastructure environment for the rapid deployment and distributed efficient execution of large-scale scientific applications. More and more scientists use workflows to build their complex applications and deploy these applications on cloud platforms for execution. However, the on-demand use of the cloud also brings many challenges to workflow scheduling in the cloud environment. On the one hand, the pay-per-use model of the cloud makes it necessary to consider the execution time and cost of applications at the same time when scheduling workflows, which increases the difficulty of s...

AI Technical Summary

Problems solved by technology

However, when faced with large-scale task requests, the Q-learning algorithm requires a large amount of data storage, and its inherent Q-value matrix dimension explosion problem will lead to high algorithm storage complexity

Based on the DQN algorithm, the value function approximation is used to solve the high-dimensional data storage problem of Q-learning, but because the fixed-dimensional environment state vector and a single type of workflow are used to train the reinforcement learning model, its generalization ability has a relatively large Limitations, it is difficult to adapt to different sizes and different types of application scheduling requirements

Based on the policy gradient network combined with the timing model, it can overcome the shortcomings of the DQN algorithm to a certain extent, but its single-strategy model shows insufficient solving ability when facing multi-objective optimization problems in complex multi-cloud scenarios.

Images