The Design And Implementation Of Scheduling Policy Of Multi-cluster Oriented Grid Computing

Grid is a distributed and parallel computing support platform, based on gridcomputing system not only enables people to gather the distributed computingresources to the formation of super computing ability, but also enable people to shareand make full use of various resources of network. Therefore, the grid is anattractive high-performance computing platforms.Platform LSF is the industry’s leading software products in grid computing field,LSF Multi-cluster can make multiple heterogeneous computers through the LAN orWAN to share computing resources and provide user a unique information andservice interface to access the resources transparently. LSF Multi-cluster finallyimplements resources sharing and co-operation with remote clusters based onspecify scheduling polices both on local and remote clusters.In this article, I extended the scheduling policies for the new system frameworkdesigned by the Platform LSF product-based scalability and performance factors inorder to provide a more efficiently, reasonable scheduling policies for LSF product.The new scheduling polices can effectively improve the operating efficiency of thejob, and maintain the load balancing among multiple clusters to meet customer needsand improve user productivity.This article designs and implements four scheduling policies according to thearchitecture of the existing product, operating mechanism, as well as the codingstandard, include job forwarded limit scheduling policy, cluster select schedulingpolicy, host select scheduling policy and job re-scheduling policy. These fourdifferent scheduling policies influence the scheduling of the job from the differentemphases can be perfectly integrated with the original system, and can also dispatchthe job to the suitable remote clusters to improve the efficiency of job execution inorder to achieve the purpose of load balancing.