Performance of redundancy methods in P2P networks under churn

摘要

Peer-to-Peer (P2P) networks organize large amounts of resources distributed across the Internet for user access. However, the availability of these resources is hindered by the participants' transient connectivity. Redundancy can be used to deal with this problem, but the redundancy repair traffic must be controlled to prevent performance degradation. In this paper, we present a comprehensive analytical formulation for erasure code redundancy methods in P2P networks, focused on fragments availability and redundancy repair cost. We use this model to evaluate the effectiveness of a proposed new redundancy scheme (PR) that outperforms other methods. Our results provide important insights for the design and construction of automated content availability mechanisms for P2P overlays under churn. Our evaluation framework allows us to determine at which point substituting a traditional erasure coding method (MDS) with other schemes is viable. For instance, e_MBR network coding performs worst than MDS for average fragment availabilities below 0.52. In addition, our experimental results indicate that adaptive repair strategies with elastic maintenance intervals can produce significant bandwidth savings compared with reactive and periodic redundancy repair strategies. In our experimental evaluation, an adaptive repair strategy consumes up to 49% less bandwidth than its periodic counterpart.