Models And Evaluation To The Availability Of Redundant Data In P2P Networks

With the rapid development of Internet technology and applications, P2P (Peer-to-Peer) network has shown great advantages in data sharing, streaming media and other application fields, and has broad application prospect. High data availability is the basis of ensuring the development and applications of P2P storage systems. Based on P2P storage system, we mainly research the data duration in terms of data redundant strategies to explore persistence and availability of redundant data.First, we propose a model to optimize the parameters of erasure codes, which is based on the P2P storage system using MDS erasure codes as the redundant strategy. This model can minimize the total system cost, and achieve data reliability.Then, we aim at three typical distributions (Exponential, Pareto, Weibull) which the session length of nodes follows to evaluate the persistence and availability of data. Meanwhile, we study that how node-join churn and erasure codes parameters have impact on data duration.After that, a model that calculates the upper bound and lower bound of mean data duration is proposed based on reliability theory about parallel system and replication redundant strategy. While the session length of single physical node follows heterogeneous exponential distributions, the synthetic session length of all nodes storing data can fit with heavy-tail distribution approximatively. Furthermore, we compare the upper bound and lower bound of mean data duration with simulation results, through traces of simulative nodes and real systems, such as Skype, Microsoft, KAD to validate the upper bound and lower bound models.Finally we summarize the research work, and prospect the future work.