Improved Optimization For Data Disaster Recovery System Over Low-bandwidth Networks

No matter LAN, MAN or WAN, the effects of the networks are irreplaceable asbridges to link primary centers and backup centers. Bottlenecks of each above ascomponents for data disaster recovery system have been definitely regarded asserious limiting factors of the integral performance improvement. Replicating thecritical data across the Internet or WAN is frequently used as backups againstdisasters, whereas the far distance makes the WAN as the bottleneck. No doubt,solution-based local replication can achieve better performance optimizations, butthe explosive growth of big data still requests better transmission efficiencies.Deduplication, compression and other similar technologies can take advantageof the time and space characteristics of data flows, and effectively reduce thenetwork load by deleting duplicate chunks or saving only the unique blocks. TheHidden Markov Model helps us hypothesize hidden sequences by using observationsequences. In this thesis, we studied techniques and its applications of existing datadisaster recovery systems on the basis of reading numerous related literatures homeand abroad. Aiming to optimize the defects and insufficiencies, we designed andimplemented a new data disaster recovery system over low-bandwidth networks.The main work of this thesis is summarized as follows:(1) Study and optimize the architecture of data disaster recovery system. Thisthesis has deeply studied the mode and the storage method of data disaster recoverysystem and mainly concerns about two issues: optimizations for architectures,modules, functions, and the approach for protecting data security.(2) Design and implement a data disaster recovery system over low-bandwidthnetworks named InfoDr system. We proposed a kind of workflow processingmechanism based on deduplication and delta compression, which makes contributionfor processes including workflow pretreatment, data segmentation, fingerprintextraction, collision detection and so on. It always helps optimize the network loadfor the primary center. Experiments showed that nfoDr system in the actualproduction environment can effectively optimize the workflow load, and itsperformance can also meet the needs of the business system.(3) By observing and analyzing InfoDr system QoS, we built a Hidden MarkovModel for low-bandwidth networks and managered to predict and evaluate the hidden state sequences from observetion sequences based on PGM. Experimentsshowed that InfoDr system can increase idle bandwidth use, alleviate the networkcongestion and guarantee the usability for multi-services.(4) Summarize the main contributions of this thesis and propose the future work.