Research On Cloud Resource Scheduling Method Based On Deep Reinforcement Learnin

At present,cloud computing is widely used in all walks of life.Its advantage is that it can provide computing services according to user needs.This new computing mode can reduce user application maintenance costs and improve user satisfaction while ensuring user service quality.Cloud service operators need efficient resource scheduling strategies to allocate their computing resources to appropriate user requests to maximize benefits.However,in the face of a growing variety of user needs,cloud computing faces challenges such as mobility factors,resource heterogeneity,user random behavior,and task dependency.Traditional job scheduling strategies,heuristic scheduling strategies,and scheduling strategies based on economic models cannot achieve a good balance of interests between users and operators,and often only make the best or best allocation at present,without considering the long-term benefits in the future,This is the key problem to be solved in cloud resource scheduling.With the continuous development of deep learning and reinforcement learning,the perceptual decision problems of various complex systems have been solved by deep reinforcement learning,and have achieved good results,which provides a new solution for cloud resource scheduling.Deep learning has strong context-aware ability,but weak decision-making ability;However,reinforcement learning is good at making decisions and has no way to deal with perception problems.Therefore,it is a new attempt to combine the two and give full play to its own advantages,and use deep reinforcement learning to solve complex problems in cloud resource scheduling.The main work of this paper is as follows:(1)Formalize the description of resources in the cloud environment,establish a system model for cloud resource scheduling,and model cloud resource scheduling using Markov decision processes.(2)Dynamic scheduling of random edge cloud environment based on deep reinforcement learning.Due to the randomness of user requests in cloud tasks,the existing heuristic and reinforcement learning-based methods lack rapid adaptability.The asynchronous-advantage-actor-critic and residual recurrent neural network are combined in the task scheduling of random edge cloud environment.The model can conduct decentralized learning,capture host and task parameters and time patterns,and quickly adapt to dynamic scenes with less data.(3)Multi-phase cloud resource scheduling based on deep reinforcement learning.In view of the dependency relationship between cloud tasks,the processing capacity and computing resources of cloud servers are different,resulting in a high rejection rate when responding to a large number of cloud tasks with dependencies.The deep reinforcement learning algorithm is introduced into the cloud resource scheduling model,and the scheduling process is divided into two stages,with the goal of reducing the rejection rate and energy consumption,to achieve a multi-stage cloud resource scheduler based on deep reinforcement learning.