Design And Implementation Of Job Scheduling Management System On Supercomputer CAE Cloud Platform

With the development of computer technology,computing power of super computer hardware in our country advances rapidly.However,the application ability of super computing is still poor.Building CAE(Computer Aided Engineering)cloud platform in super computer is an effective solution.Job management system is the core of the platform system.CAE jobs have strong dependence between them.When a task is in the implementation,it needs to wait for the completion of its precursor task,which is not the same processor.Because of the communication between tasks,the work time is long.Thus,the leisure time between the task and the precursor task on the same processor is longer.If the idle time is not reasonable used,the system will increase the execution power consumption of processor.The purpose of this paper is to design and realize the CAE cloud platform job scheduling management system.Based on dependent task duplication and processor combination,we put forward a kind of leisure time energy-efficient optimizing scheduling strategy to reduce the energy consumption of job scheduling operation.By the experimental simulation and real example test,we have verified the effectiveness of the strategy.In this paper,the main work includes two points as follows:(1)A CAE cloud platform job scheduling management system has been designed and implemented based on super computer,through this platform,the user can access to the super computer CAE resources in WEB way at anytime,anywhere,all they need to do is to pay on demand,thus full enhancing the super computer powerful computing ability and greatly reducing the use cost.However,the CAE jobs usually have the very big dependence with long job completion cycle,causing longer waiting time between the communication between of dependent tasks,thereby increasing the same processor on the spare time between two tasks.The long job completion cycle and idle time increase energy consumption of the job scheduling management system.(2)A kind of free time to optimize energy saving scheduling strategy based on dependent tasks has been put forward,considering the energy consumption of the CAE job scheduling problem.Firstly,in order to reduce the task scheduling length,copy tasks scheduling algorithm on the critical path to the non-critical path based on task duplication;Then merge tasks on less running task processor to more running task processors,so as to reduce nodes energy consumption;Finally,use step-down Frequency reduction DVFS(Dynamic Voltageand Frequency Scaling)strategies to reduce energy consumption on the combine processor with free time.Experiments carried out and the proposed strategies are integrated into the job management system on CloudSim simulation platform,through the testing of the actual example,the experiment and test results show that under the condition of invariable scheduling length,the algorithm can effectively reduce the energy consumption in the process of job scheduling operation.In this paper,the energy-saving strategy results show that the proposed strategy can better reduce the energy consumption in the process of job scheduling execution.In addition,because of using cloud technology,users can have on-demand access to supercompute CAE resource anytime and anywhere,which greatly improves the user's work efficiency and has great application value.