Decentralized Integration of Task Scheduling with Replica Placement

Data Grid integrates graphically distributed resources for solving data sensitive scientific applications. The main issues in data grid are task scheduling and data management. As data grid spans multiple organization areas, it makes centralized resource management difficult. Therefore, it is necessary to study decentralized resource management. In this paper, a decentralized architecture of integration of task scheduling with replication placement is put forward in advance. Based on this architecture, a game theory based decentralized replication placement model and related algorithm, best-reply algorithm, were proposed. At last, four compositions of task scheduling and replica placement algorithms were compared by simulations in terms of average job completion time and average network load. The result shows that although the integration of decentralized online task scheduling algorithm with best-reply algorithm, against centralized integration algorithms, is a little worse in average job completion time, its average network load changes a little and it can be substituted for the centralized integration algorithms whatever the size of disk space of storage resources is.