Research And Implementation Of Grid Job Management System

The Grid aims at realizing resource sharing and cooperation on the Network Environment, as well as preventing Information and Resource Island. Since 1990's, America and some European countries have already made progress on the basic research of the Grid Technology and on the system development. At the same time, some research are currently running in China. Grid Job Management is an important aspect of the Grid technology.With the development of the Grid technology, especially the proposal of the OGSA, it indicates that it is the trend to provide standard interfaces for Grid resources using service. China National Grid strictly follows OGSA specification, which encapsulate all the underlying resources in service. How to run service-oriented job management and shield the access of underlying service resource in order to optimize Grid service resource and provide well support for the Grid application is one of the most important issues of the Grid technology.The background of this subject is China National Grid (CNGrid). A new service-oriented Job Management System has been designed and implemented, which provides job management service for the CNGrid, such as the parsing of job request, resource selection and global scheduling. This system adapts Grid service mechanism and describe job request in XML format. It can choose the optimal Grid resource dynamically according to the user's request and it can achieve load-balance at the same time. It virtualizes the Grid resources and implements access of the layer of virtual resource. It can provide resource selection interfaces of custom policy and global scheduling interfaces, which can enable the developer to configure new scheduling policy according to the new requirement. This system is robus, scalable and acts as the proxy for the access of Grid resources, while providing location-independent and session-support job management service interfaces. According to the experiment, this system could improve the performance of the job execution and decrease the average overload of a single job execution obviously.