The Design And Implementation Of Mechanism For Data Availability Management In Hybrid Backup System

Availability is one of the most important properties for backup system. In traditionalbackup system of C/S architecture, availability is achieved through the redundancy ofbackup data. In P2P hybrid system, data is distributed between storage server and P2Psystem. Current backup system monitors data stored in P2P system through the backupclients or peers in P2P system, and performs replicate operation when the remaining datais less than threshold. How to repair the loss data fast and how to make the communicationoverhead minimize are important problem of the availability management in hybridarchitecture backup system.This availability maintenance strategy based on lease can be applied to P2P hybridarchitecture system. We adopt the lazy repair strategy. DIRD manages the metadata oferasure code fragment centralized and selects one peer from all peers which is holdingerasure code fragment belongs to the same backup object as lease peer. Lease peerestablishes a TCP connection with all peers one by one to perform monitoring accordingto the metadata stored in the lease. When lease peer probes that the rest of erasure codefragments below the threshold, it will send message to DIRD to request repair operation.The online status of lease peers is monitored by DIRD. DIRD selects target peersaccording to the principle of load balancing, and performs replication operation accordingto repair strategy after receiving the requisition for repairing. DIRD configures thethreshold according to dynamic threshold policy, so as to realize the “flowing” of theoverhead for availability management between storage and P2P network.According to the experiment under a10peers P2P network, this mechanism foravailability management works well and it repairs loss data very fast. According to thesimulation under a5500peers P2P network, its communication overhead is very low andthe effect for load balancing is significant.