Coding And Scheduling Considerations For Peer-To-Peer Storage Backup Systems

Peer-to-peer storage- and in particular backup-system architectures have recently attracted much interest due to their use of "free" resources, with disk spindles and communication bandwidth being at least as important as storage space. This paper complements most of the works on this topic, whose focus was on metadata, security, locating the stored data, etc., by focusing on the data itself. It offers important design considerations and insights pertaining to the composition of erasure-correction code (ECC) groups, their size and the level of redundancy. Dynamic issues such as the co-scheduling of the concurrent reconstruction of multiple ECC groups are also explored. Finally, we identify an interesting natural match between asymmetric communication bandwidth (e.g., ADSL) and a hierarchical reconstruction architecture aimed at alleviating bottlenecks at the reconstructing node.