Replica Technology Research In High Performance Computing Environment

The Computing grid is the new high-performance computing environment, which is also a grid technology with an emphasis on computation. Resources are shared in the computing grid to improve the computing capability. In high performance computing environment, data resources, the key department of all computing resources, are large-scale, data-intensive, geographically distributed and accessed or operated frequently by computing applications. Huge distributed data is the most important to manage in all computing resources. Therefore, the replica technology has been widely used in the computing grid. By creating replicas locally of the remote file, it is reduced to access times from remote machines, balance the work load and then optimize the whole network's performance. It has been a challenging problem to find where, when and choose which file to create or delete replicas. Based on the University Computing Grid (UCGrid), this thesis has studied mainly on the dynamic management of replica, suggests a new optimization algorithm. And several algorithms are simulated and analyzed with simulator. On the other side, the new algorithm is also realized in the real grid environment.In this paper, the development and effect on replica technology of the grid infrastructure are studied. The foreign and domestic research situation of replica is introduced. Since then the method of replica management in Globus system and basic replica components are studied. Other algorithms'theories and features are discussed with an emphasis on the factors about algorithms'capability. Base on the character that the cascading strategy is designed for layer grid architecture and other strategies'advantages, the new replication algorithm (Min-Transport Cost and Aggregate strategy MTCA), which is in order to reduce the whole transport cost, is designed. The simulator for algorithms is studied and modified in order to compare different algorithms. After the necessary preparations such as adding the Cascading and MTCA algorithms and modifying the simulator's replica management model, the experiments and analyses are carried out successfully.Finally based on the extended replica strategy, how to develop an application for replica management is researched. The replica management and scheduler system is designed and implemented according to distributed features of the Web Service technology in Globus. The replica management and scheduler system adopts the new replica strategy (MTCA) to manage replicas dynamically in grid environment. A portal for the replica management and scheduler system is realized too. Through the portal, grid users can operate replicas much more continently.With many experiments and analyses, it has been proved that the min-transport cost and aggregate strategy (MTCA) can reduce the mean job time, reduce network and storage resources usages obviously, The strategy also can be applied to Globus grid environment easily, and improve the efficiency of replica management effectively in the grid environment.