Research On Multi-Core Task Scheduling Method Based On DQN

With the increasingly complex demand of multi media and scientific computing,it is urgent for computers to have more powerful computing power.Due to the limita tion of semiconductor technology,it is difficult to meet the increasing demand for processing performance improvemen t by simply improving the working frequency of the proces sor.Therefore,the multi-core processor technology emerges as the times require.Through the parallel processing of m ultiple processor cores,the processing performance can be i mproved rapidly.With the development of multi-core technology,multi-core task scheduling refers to a set of tasks with certain constraints and dependencies,which are allocated on the multi-core according to certain methods to achieve the highest efficiency of computing resources as a whole.First of all,through the analysis of the multi-core task scheduling problem,this paper designs the multi-core task scheduling model from three aspects of processor model,task model and scheduling model,and points out that themulti-core task scheduling problem is a NP complete problem,which can not find the optimal solution in the effective time,while the Q-learning algorithm does not need to consider the environment model,which avoids the problem of too large scale The impact of multi-core task scheduling.Secondly,due to the large scale of action space and state space of multi task scheduling problem,this paper adopts deep neural network,introduces the mechanism of experience playback and target network,designs dqn algorithm.Based on the Markov property of Q-learning algorithm,each "state action" is regarded as an independent training sample.By dqn algorithm,the complete action track is divided into several relatively independent "state action" modules,and a sample database is established.In each iteration,new samples are added to the sample database.When the number of samples in the sample database reaches a certain number,some samples are taken from the sample database for strategy evaluation and improvement.Finally,500 DAG graphs of random task sets are randomly generated from two indexes of makespan and SLR.The comparison between heft,ppeft,HLD and dqn algorithm is carried out.The experimental results show that dqn algorithm studied in this paper has better performance of multi-core task scheduling.