Reliability estimation of learning based mobile agent system in MANET

Mobile agents have found wide use in wireless and mobile network applications. This paradigm provides a uniform approach to many problems and is customizable for the user. But dependability issues like reliability and availabilityneed to be addressed before agents can be commercially used in Mobile Adhoc Network (MANET). Since MANET provides a resource constrained dynamic environment, it introduces new aspects to dependability thereby affecting reliability of the services provided by the mobile agent based system (MAS) deployed in MANET. In this paper we show that MAS on MANET can be made more reliable if the agents are intelligent enough to share information and learn (reinforcement learning) about the underlying MANET conditions dynamically. A protocol is proposed to estimate reliability of learning based agents. The MAS consists of independent agent groups pertaining to some application for which these are deployed. The results achieved shows that learning yields better performance and improved reliability in MANET. As a MANET grows the effect is even more apt.