CNP based negotiation protocol for multi-agent systems

Multi-agent negotiation and task allocation is a fundamental research issue. Various negotiation protocols have been developed. One of them is the well known CNP protocol which is defined by Smith and Davis. It is used for decentralized task allocation and based on the notion of call for bids. There are manager and contractor agents. The system is dynamic and each agent on the network can join or leave the system at different times. The CNP protocol has some limitations. Firstly, in a distributed environment, several managers can concurrently call for bids, so an agent may have to manage several negotiation processes in parallel in order to reduce the length of its negotiation processes. Some applications of the CNP force the contractors to sequence their negotiation processes. Thus, when there are several managers that call for bids, contractors cannot give proposals in parallel. Thus negotiation processes take longer, and meanwhile, contractors may miss some contracts. Secondly, CNP-based applications enable the agent to break its commitments when an agent receives an offer for a better task in comparison with those for which the agent is committed. If a manager comes with a better task, contractors may break their commitments. This is not a good solution because it makes managers call again for bids to find new contractors for their tasks. In case of agent failures, negotiation processes may be blocked. This study describes the design and implementation of an agent based execution environment where agents follow the extended CNP negotiation protocol defined by S.Aknine, S. Pinson, M.Shakun during negotiations. An application based on a transportation problem has been developed to assess the system. In this study, JADE is used to implement the multi-agent system. Various experimental studies have been performed for different numbers of manager and contractor agents and the results are compared. The success key is the negotiation time and the number of the steps applied until the negotiation process is completed. Increased in the number of the contractor and manager agents cause the negotiation time take longer. Better results have been observed when the best offer is also sent by the PreReject message compared to not being sent, especially for more complex and crowded agent systems.