High-performance Computing Cluster Management System And Job Scheduling Technology Research And Realization

Currently, network-based computer technology makes a great push towards the development and wide application of cluster system. High-performence workstations or PC connected in a certain structure by high-speed network into cluster, then scheduled, so as to get the similar functions of large machine and parallel machine with a very small amount of cost. The software system used to manage these workstations and PC are the targeted Cluster Management System in this thesis, and job scheduling technology is one of the key technologies of cluster management system. New-generation Cluster Management System's study and application technology is regarded as the mainstream study approach in the field of high-performence computer over the past few years. Many civilly or commercially used software and related products have been promoted.Our project is aiming to develop a comprehensive Cluster Management System named Origin CMS (OCMS for short), expecting that the system owns such characteristics as distributing operation, resources share, load balancing, as well as fine operation, high reliability, good extensibility and preferably adapting the special requirements of climate numerical nalue forecast. The author' s major research achievements are as followed.Firstly, in terms of Cluster Management System's system structure, the author puts forward a new kind of system framework for cluster management and job scheduling with three-level B/S structure, featuring non-platform and remote management.Secondly, on the basis of OCMS ' s application, conduct deep and comprehensive research in complex job, job scheduling, load balancing, network communication, easy-operating etc.Thirdly, job schedule is a key element in realizing fairly dealing with client operation, shortening system feedback period, so as to improve system's performence. Considering the important function in Cluster Management System,the author is conducting cluster system's job scheduling research, raising an applicable, stable and reliable job scheduling method, which proves a very effective method to the core issue in OCMS.Fourthly, resolve the problem that the job scheduling and real time feature adapt for the requirements of climate numerical nalue forecast job.On the basis of deeply research the principle and architecture of Cluster and Cluster Management System, author design and implement the OCMS and job scheduling subsystem which is the core part of OCMS.