Research On Data Replication For Mobile Ad Hoc Networks

A mobile ad hoc network (MANET) is a special wireless network that is self-organized by mobile nodes communicating with distributed protocols. MANET can work in absence of fixed infrastructures, therefore it can be used in many situations where temporary network connectivity is required, for example in battlefields and in the disaster recovery.The data replication technology is crucial to improve data availability and data access performance. However, data management is more complicated in the MANET that has such basic characteristics as self-organization, mobility, multi-hop and limited resource. Therefore the research on data replication for MANET is quite challenging. To address the problems that are caused by the nature of MANET to data replication, a Clustering-based Data Replication Protocol (CDRP) framework is proposed in this paper.The CDRP includes clustering algorithm, metadata management protocol, replica management protocol and probabilistic reliable dissemination protocol. The CDRP mainly aims to improve the data availability, and then to reduce communication cost of data access to improve data access performance, and to reduce the network load to improve the scalability of the protocol. In this paper, we mainly study the clustering algorithm, the replica allocation algorithm, and probabilistic reliable dissemination protocol in the CDRP.The foundation of the CDRP is the a-Stable Path-based Clustering Algorithm (a-SPCA). The purpose of the a-SPCA algorithm is to predict network partitioning by clustering. The a-SPCA algorithm predicts the link availability between two neighbor nodes depending on the current locations of these nodes and the mobility mode, and consequently predicts the path availability between pair of nodes. Based on the path availability, the a-SPCA algorithm dynamically organizes mobile nodes into clusters in which the path availability between any pair of nodes is greater than a certain threshold. Since the criteria for cluster organization depends directly upon the path availability, the structure of the cluster is adaptive with respect to node mobility and can efficiently predict the network partitioning. Theoretic analysis shows that our algorithm has proper complexity. Simulation results show that the clusters created by the a-SPCA algorithm have desirable properties.The replica management schedules the number and the location of replicas. On one hand, mobile nodes move freely and link disconnections occur frequently. This may cause frequent network partitioning, i.e. the network is separated into completely disconnected partition. Network partitioning can cause sudden and severe disruptions to ongoing data access, and accordingly reduces the data availability. To improve the data availability, we propose a Clustering-based Data Replica Allocation algorithm (CDRA). The CDRA algorithm is based on the a-SPCA algorithm and dynamically schedules data replicas in every cluster. Each cluster head counts the access frequency from the whole intra-cluster nodes to each data itemand decides which data item to be replicated in this cluster. The CDRA algorithm re-schedules replicas while nodes join or leave the cluster. Predicting the network partitioning by clustering and replicating data before network partitioning, the CDRA algorithm improves data availability. On the other hand, the communication cost has become a prominent factor influencing the performance of replica allocation in the MANET environment. To improve the performance of data access, a dynamic adaptive replica allocation algorithm based on the stable neighbor (EARAMSN) is proposed. When changes occur in the network topology or the access requests of the data item from intra-cluster nodes, the EARAM_SN algorithm dynamically adjusts the replica allocation in a cluster to minimize the communication cost of the data access and thus improves the performance of the CDRP.The PercoLation-based Probabilistic reliable Dissemination protocol {PLPD) is used to disseminate data lookup lists and update messages of data replica among clusters to guarantee the consistency of the lookup lists and replicas among cluster heads. Different from other probabilistic reliable dissemination protocols, the PLPD protocol is a network topology-aware group communication protocol. It does not require flooding the network with control messages to manage group membership. In PLPD, each group member only maintains a partial view of other members in its neighborhood and disseminates received data messages with probability p to a subset of the view. Because both the membership management and the selection of message dissemination destinations in PLPD are local and are not related to remote nodes, the network load is decreased. We model the dissemination process of the PLPD protocol with the percolation theory. It is proved that the PLPD protocol can achieve a probability with 1 for all group members to receive every message when p is greater than a certain critical threshold. The simulation results show that the PLPD protocol effectively reduces the network load while providing high probabilistic reliability, and scales well to large system sizes.The research on data replication for the MANET will greatly promote the application of the MANET. However, relevant researches in this field are just beginning. Many efforts still need to be made to realize its future applications.