Research On Combined Scheduling Strategy Of Multiple Supercomputer Centers

In recent years,China’s supercomputing business has ushered in rapid development,and supercomputers have become an important tool for promoting the development of scientific and technological innovation.However,users’ demand for supercomputing computing power is growing rapidly,and the increasing emergence of new types of computing services has also put forward stringent requirements on computing scale,latency,flexibility and other aspects.The emergence of multiple supercomputing center interconnections can solve the above problems.Multi-supercomputing center interconnection can integrate the computing resources and storage resources of supercomputing centers across geographical regions,and can efficiently organize multiple supercomputers across geographical regions to provide users with diversified computing power services,which is an important way to support high-quality computing services.However,how to efficiently match the computing nodes’ arithmetic power with users’ jobs in the multi-supercomputing center interconnection system is a problem that needs to be solved urgently.Therefore,this thesis focuses on the research of multi-supercomputing center interconnection,and discusses the scheduling strategy of arithmetic power combination to optimize the completion time of the overall job according to the demands and characteristics of different user jobs.The main research contents and contributions of this thesis are as follows:(1)For the problem of processing computationally intensive jobs in cross-region supercomputing centers,this thesis proposes an ant colony algorithm with improved knowledge.The algorithm solves the problem that some supercomputing computing power is difficult to meet the user’s jobs with strict deadline requirements.In this thesis,we first mathematically model the combined scheduling of the computing power of multiple supercomputing centers,and construct an overall job completion time minimization model under the constraints of the job deadline of multiple supercomputing centers and the storage resources of computing nodes.Then an improved ant colony algorithm is designed to solve this problem.The algorithm is based on the classical ant colony algorithm,and the knowledge body is designed to influence the ant movement direction by modifying the state transfer probability formula and then affecting the ant movement direction.Simulation results show that the improved algorithm can improve the accuracy of matching supercomputing arithmetic power and jobs,and can quickly schedule the arithmetic resources of multiple supercomputing centers,which improves the resource utilization of multiple supercomputing centers.(2)For the problem of processing data-intensive jobs in cross-region supercomputing centers,this thesis proposes a heuristic arithmetic power combination scheduling algorithm.The core idea of the algorithm is to combine data locality with load balancing of supercomputing centers,to reduce unnecessary data movement by transmitting jobs with smaller data volume through the network,and to support data processing in non-local supercomputing centers.In this thesis,we first split the job into multiple sub-jobs,formulate the problem of minimizing the completion time of data-intensive jobs in multiple supercomputing centers as a linear programming,and design a heuristic algorithm.The algorithm first orders the jobs according to the completion time,and then prioritizes the placement of jobs with minimum completion time,while taking into account the load balancing of jobs in supercomputing.Simulation results show that the algorithm provides a combined scheduling strategy that significantly reduces the overall job completion time with high accuracy and robustness.