An Agent-based adaptive task-scheduling model for peer-to-peer computational grids

This paper presents an agent-based adaptive task-scheduling model for pure P2P computational grids, in which the task-scheduling mechanism is recursive, dependable and purely decentralized. The main idea is that the application provides the task decomposing method and decomposition is triggered by the platform once parallel running is possible. Mobile agents are used to carry tasks and results moving from one node to another. Each node has a manager agent administrating other agents to distribute and schedule tasks. From the preliminary testing results, it can be seen that the model can provide a low-cost large-scale computing platform on the pure P2P architecture. It avoids the disadvantages of hybrid-P2P architectures and can adapt to the applications that can be decomposed into independent coarse-grained subtasks without large amounts of data.