Performance enhancement of a contract net protocol based system through instance-based learning

The contract net protocol (CNP) is a widely used coordination mechanism in multiagent systems. It has a lot of communication overhead due to the broadcast of the task announcements. The performance of the CNP degrades drastically when the number of communicating agents and the number of tasks announced increases. Hence, it has problems of scalability. In order to overcome this limitation, an instance-based learning (IBL) mechanism is designed that uses previously stored instances in order to select a target agent. This avoids the expensive bidding process. The scheme is implemented in a simulated distributed hospital system where the CNP is used for resource sharing across hospitals. Experimental results demonstrate that with the incorporation of the IBL, the system performance improves significantly. The system is better scalable with respect to the number of tasks.