Research On Job Assignment Problem Of Server Cluster Based On PS Rules

There are tens of thousands of data centers in the world.With the development of technology and the increasing demand for data services,the number of data centers is increasing every year.While bringing convenience to our work and life,data centers consume a lot of electricity and bring a lot of carbon pollution every year.The purpose of this dissertation is to design the optimal assignment method for the incoming job requests in the server farm,so as to improve the energy efficiency of the system to achieve energy saving and ensure the efficient completion of the job.Job assignment mainly studies how to allocate appropriate server nodes for the jobs submitted by users.An excellent assignment algorithm can not only guarantee the stability and robustness of the server farm operating environment,but also shorten the mean service time of jobs,improve service efficiency and improve user satisfaction.The goal of this optimization is to improve the energy efficiency of the system and reduce the mean service time.We consider a large,heterogeneous server farm with a realistic dimension,where the servers in the server farm are divided into a certain number of server groups.Isomorphic between the same group of servers and heterogeneous among the different groups of servers.All the servers handle the arriving jobs with the Processor Sharing(PS,processor-sharing)discipline,and each server has a limited number of buffers.We model this system as M/M/K/S/PS multi-queue and multi-server queuing system.The research focus is to design the job assignment method to improve the energy efficiency of the system and reduce the mean service time of the job.For this reason,we propose an OAIP job allocation method based on the global consideration of the system.The OAIP method focuses on reducing the energy consumption of the supporting processing service rate(that is,improving energy efficiency)and improving the processing efficiency of the job.Before studying the assignment method of M/M/K/S/PS model,we first studied the M/M/1/K/PS model with only one server.According to the knowledge of relevant queuing theory,we derived the mathematical expressions of five performance metrics of M/M/1/K/PS model by mathematical analysis,and compared the analysis results with the actual simulation results,so as to verify the correctness of the simulation results of M/M/1/K/PS model,which laid a foundation for the simulation of M/M/K/S/PS model.Before designing OAIP method,onlythree servers in a simple model of M/M/K/K assignment method is studied.For this,we design the RMAIP method,using mathematical analysis method the RMAIP method is obtained with MAIP expression method of performance measurement,and also has carried on the simulation and comparison,the final result shows that the simulation results accord with mathematical analysis results.RMAIP method is better than MAIP in terms of energy efficiency and average service time in small-scale server farm.The design basis of RMAIP is closely related to the OAIP method.The completion of the analysis and simulation of RMAIP method lays a foundation for the design and analysis of the following OAIP methods.Based on M/M/K/S/PS OAIP model simulation and design methods,we get the results,with the latest,progressive optimal MAIP method contrast,observed OAIP method can improve the energy efficiency of the system and reduce the mean service time,in the performance of the energy efficiency performance and MAIP method is very close and the gap is very small,the most important thing is OAIP method of energy saving at the same time can greatly reduce the mean service time,that means can greatly improving the efficiency of the operation of the service.The experiment shows that the OAIP method is not sensitive to job size distribution,so the OAIP method we designed is of great practical value.