Peer-to-Peer Scheduling System with Scalable Information Sharing Protocol

In traditional job scheduling systems for the grid, a single or a few machines handle information of all computing resources and scheduling tasks. This centralized approach is not scalable, since it introduces single point of failure and bottleneck. Some decentralized scheduling systems have been proposed to improve scalability. They avoid concentration of scheduling costs by broadcasting job execution requests in a peer-to-peer style. However, resource utilization tends to be low, because most of them are not aware of the dynamic states of computing resources before throwing execution requests. This paper introduces a decentralized scheduling system that improves resource utilization by using a Gossip-based multicast protocol. With this protocol, peers can gather information of each other efficiently and schedule jobs individually. The simulation shows that our system is scalable and it handles many jobs efficiently in large scale grid environments