Modeling And Analysis Of P2P Storage Systems With Replication

With the rapid development of massive storage system, there are more and morestorage applications based on P2P network. How to manage the data replicas effectivelyand make full use of the limited network resources has become a research focus. In thispaper, taking into account the peer dynamics in the P2P storage system, based on thereplica repair policy and the data access process, the mathematical models are constructedto evaluate and optimize the P2P storage system.Firstly, based on the working principle of the pure P2P storage system with eagerreplica repair policy, two possible states of a replica are defined, i.e., available state andbeing repaired state. By considering both the data access procedure and the replica repairprocess, a two-dimensional Markov chain model is established. Based on the workingprinciple of the pure P2P storage system with lazy replica repair policy, three possiblestates of a replica are defined, namely, available state, waiting for repair state, and beingrepaired state. By considering the data access procedure, the replica repair waiting processand the replica repair process, a three-dimensional Markov chain model is built.Secondly, based on the working principle of the Peer-Assisted cloud storage system,with lazy replica repair policy, taking into account the data access procedure in cloudcenter, the data access procedure and the replica repair process in P2P storage layer, athree-dimensional Markov chain model is constructed.Thirdly, the request arrival, the data access time, the replica life-time and the replicarepair time are assumed to follow exponential distributions. By using the matrix-geometricmethod, the system models are analyzed in steady-state. Accordingly, the performancemeasures in terms of the data availability, the average access latency and the replica repairrate are given.Finally, numerical results with analysis and simulation are provided to demonstratehow system parameters, such as the replica number and the replica repair threshold for adata, influence the system performance. Moreover, benefits functions are developed tooptimize the replica number and the repair trigger threshold in P2P storage system withreplication, respectively.