| Since the9/11events in the United States, governments around the world havepaid more and more attention to improve the service emergency response capabilitiesfor information systems; meanwhile, service emergency response/disaster recoverymechanisms have been built in many key industries and areas, such as banking,insurance, etc.. In service emergency response system, protection for the key businessdata is one of the key steps, which is strongly related to the success or failure of theentire service emergency response process. However, traditional data protectiontechnology is difficult to meet the high reliability, high availability and high real-timedemands of the service emergency response.Towards many challenges for the data protection design oriented serviceemergency response, providing highly reliable, high real-time data protection capabilityand ensuring high availability for service application, this paper deeply researches thekey technologies of multi-versioning data protection model, data protection mechanismon file system level and block level, data de-duplication mechanism. The maincontributions of this paper are listed as follows:1. A multi-versioning data protection model oriented service emergency responsehas been proposed in this paper. By analyzing data replication frequency, versionrecovery time, data storage space overhead and data redundancy, which affecting thedata protection efficiency and time and storage overhead, from the aspects of dataversion generation, version management and version recovery, mathematical descriptionfor multi-versioning data protection mechanism is done to propose a multi-versioningdata protection model oriented service emergency response. Based on the traceinformation of the online transaction processing system, the relevant characteristics ofthe model have been simulated in this paper. The model can guide the optimizationdesign of multi-versioning data protection on the file system level and block level, andgreatly improve the RPO and RTO indicators on the basis of comprehensive balance onmulti factors.2. An efficient file system level multi-versioning data protection mechanism(VerFS) has been proposed in this paper. Aimed at low efficiency of data recovery inthe traditional method of direct disk data copy, fast recovery the file data by re-linkingthe logical blocks in the file system level, and further reduce the time cost of datarecovery process combined with optimization design on disk write operationperformance. Compared with the traditional multi-versioning data protectionmechanism (COW) on file granularity, the storage overhead in VerFS is about23%to51%of the COW mechanism. The RPO value in VerFS can achieve minute level; theRTO value can reach the second level. 3. An efficient and reliable block-level multi-versioning continuous data protectionmechanism (VerCDP) has been proposed in this paper. In traditional TRAP mechanism,the data validation log chain is too long, results the low recovery performance andfailure of the recovery process. A snapshot of the current data block B is beendynamically inserted to the parity log chain of the original TRAP mechanism, and indynamic time interval, VerCDP storage the compressed results of the data change. Byincreasing some backup storage space, the VerCDP can achieve the efficient datarecovery time. Analysis showed that: compared with the TRAP-4mechanism, VerCDPincreases0.83times to2times log storage; compared with the traditional copy-on-writedata protection mechanism (COW), the storage overhead of VerCDP is about12%to42%of the COW mechanism. The recovery time of VerCDP is only about10%of theST-CDP mechanism. Meanwhile, the data recovery reliability in VerCDP has beengreatly improved.4. An efficient VerFS-oriented de-duplication mechanism (DeDupVerFS) has beenproposed in this paper. Using the pattern of hierarchical data organization andmanagement and the characteristics of metadata memory-resident in VerFS, ahierarchical detection mechanism is designed for the same data. Further, based on themechanism, the detection optimization techniques are studied based on the hashfingerprint prefix, and the time complexity of the same data detection is reduced toO(1).Experimental results show that: while the average system I/O response time is not beenimpacted significantly, the DeDupVerFS can save6.8%to18.6%of the data storageoverhead.5. Under the guidance of multi-versioning data protection model, integratingthese key technologies, a protection prototype system has been designed andimplemented for these technologies. It not only verifies the correctness and validity ofthese technologies, but also can be directly applied to the data service system which hasthe emergency response needs.Under the guidance of multi-versioning data protection model, focusing onimproving the RPO and RTO indicators, an efficient and reliable data protectionmechanism within the user tolerated increasing of storage overhead is studied in thepaper. Related theoretical analysis and experimental results show that our method hassome theoretical significance and practical value. |
PDF Full Text Request