Self-organizing communication-aware resource management for scheduling in grid environment

Rapid advances in network and computer technologies are making networked computers, organized in the form of grid, an appealing vehicle for cost-effective parallel computing. But how to handle efficiently the communications in scheduling is still a main obstacle to using these resources. In this paper, we tackle this problem by partitioning resources into groups in a parallel and distributed fashion. Resources with good communication performance to each other are clustered into a same group. Based on our observation that communication latencies between adjacent resources are much less than those between non-adjacent ones with high possibility, flooding with a small TTL (time-to-live) can inherently exploit the proximity property between resources, which improves greatly the efficiency of our partitioning work. Our distributed resource management method can fit well for environments with large-scale resources such as grid