Replication with state using the self-organizing map neural network

Most architecture of mobile ad hoc network is in the form of decentralized, self-configuring and dynamic topologies. Nodes are mobile in network. The mobility of node in network is common problem in peer-to-peer technology. Object replication is one of the techniques applied in order to share objects in the mobile peer-to-peer environment. Predicting the estimated time for the node to exit is a great opportunity to improve the efficiency of search algorithm. Even if the object owner had departed from network, the shared objects are still available at all times. The goal of the technique is to maintain a number of object replicas over the time before a node exits in decentralized and unstructured environment. That is the reason why there is a need to replicate objects based on the predicted condition of nodes that are about to depart from the network, which is necessary. This paper proposes a modified form of random replication of data within a mobile peer-to-peer network based on predicting condition for a mobile node to replicate the object from it. It uses the unsupervised learning neural networks called the Self-Organizing Map by classifying the input attributes of each node and providing a training set - serving as a basis of identifying the nodes´ current state. Existing algorithm shows significant results such as reducing data traffic, load balancing, and decrease query latency. The preliminary results of the proposed scheme had level into the existing algorithm because of its node mobility prediction condition capability as the significant feature to replication. To test the functionality of the technique, a simulation was developed in a multi-agent based modelling environment called the NetLogo and observations are compared with the proposed scheme with the existing algorithm.