Research On Dynamic And Optimal Scheduling Mechanism For Processes In Distributed Environment

Process scheduling is a crucial scheduling strategy in distributed computing platforms,which not only directly affects user experience,but also relates to the platform’s operation cost.However,the process scheduling problem is proved to be attributed to the NP-hard problem in distributed environments.Many scholars have conducted in-depth research on this popular problem.However,the mainstream algorithms currently designed to solve the scheduling problem under distributed platforms ignore the dynamic nature of the environment during process execution in distributed platforms,and cannot respond and make adjustments to the algorithm-generated scheduling scheme in a timely manner when the environment changes.For example,when the node load changes or the node goes down due to the change of data volume,the previous scheduling scheme may be more costly or directly unavailable.To address this problem,a dynamic scheduling problem that can change and adjust the scheduling scheme according to the dynamic changes of the environment during the scheduling process is urgently needed to reduce the total consumption during the process execution.The main findings of this paper are as follows:(1)To address the uncertainty caused by dynamic changes in the process of continuous process scheduling,a process dynamic scheduling algorithm DNSGA-Ⅱ-Rn is proposed,which uses random additions to the population to increase the diversity of the population and expand the search space of the population when it detects changes in the environment to a certain extent,effectively solving the problem of population diversity.(2)To further improve the effectiveness of the scheduling scheme generated by the DNSGA-Ⅱ-Rn algorithm,the DNSGA-Ⅱ-T algorithm is proposed for data load changes and physical node downtime,and this algorithm makes different responses to these two different changes respectively.When the load changes continuously,the data of load changes in the past continuous environments are collected,and based on these data,the Transformer model is trained and learns the law of load changes,predicts the locally optimal scheduling scheme in the next environment,and adds it to the population,so as to purposely expand the search scope,improve the population quality while solving the lack of population diversity,and finally achieve the optimal scheduling scheme the purpose of optimizing the quality of the scheduling scheme.(3)An experimental validation prototype system based on the DNSGA-Ⅱ-T algorithm is designed and implemented,providing visual operation and display interfaces for model training,algorithm parameter setting,scheduling solution generation,and scheduling solution saving.