Model and Evaluation of Redundant Data Organization for High Availability in Structured Overlay Networks

The paper presents a hierarchical organization of two storage redundancy schemes to improve data availability caused by churn in structured overlay networks. To mask or hide the high churn from the portion of short-lived but frequent churn peers and permanent failure peers, we use replication among the nodes in a certain interval that can be considered as a virtual node. Then a set of virtual nodes that cooperatively provide guaranteed over the networks with erasure coding. We present the analysis of the behaviors of single one virtual node and cluster of virtual nodes. According to the stochastic models of the behaviors, the data availability under churn is presented. We also give some quantitative analysis based on one empirical trace dataset.