The Design And Implementation Of Scheduling Module Of Short Jobs In Grid

By using the internet, organizing computers in different geographical positions into a "virtual super computer", this method of computing is called grid computing. Grid computing lets people to make full use of all kinds of resources of network, coordinate jobs running in network and provide users with a unique information and service interfaces.In this dissertation, based on grid computing product called LSF which developed by Platform corporation, I raise a solution to solve the shortcomings that it has a bad performance when scheduling for short jobs, in the same time design and implement independent short jobs'module. It effectively controls short jobs's scheduling problems in the process of high performance. It makes the product in large numbers of short jobs circumstance more effective and reasonable scheduling. In this dissertation, contents of design and implementation are according to the original concept of grid computing products, and make full use of existing products and existing product framework. I developed an indenpentent module outside of the original product, which makes the load of the dispatch process decreased. Work covers designing the framework of the short jobs scheduler module and the interaction with the original system, designing and implementation the process of the jobs schedulering and resource allocation.In this dissertation, the main problems and technical key points, including the short jobs optimization problem, how to control and reduce the concurrency race condition, fault tolerance based on checkpoint, re-run task machnism and controlling of resource allocation and collection.Through the dissertation, I fully integrate short jobs scheduling module into the whole system in the end. It raises the performance of the scheduler when allocates the resources reasonably and has some measures of fault-tolerant.