Font Size: a A A

Research On Block Based Data Protection Technology

Posted on:2014-08-07Degree:DoctorType:Dissertation
Country:ChinaCandidate:J YangFull Text:PDF
GTID:1228330425973317Subject:Computer system architecture
Abstract/Summary:PDF Full Text Request
As information technologies are widely used in daily life, data loss often result in huge losses. The main reasons of data lost can be divided into two categories. One is the system’s hardware and software failures, the other is man’s wrong operation or illegal modification. System’s unexpected shutdown will direct result in cache’s data loss. Even cache using non-volatile memory, its data will also unusable due to the mapping table and status information missing. To avoid these phenomena described above, the storage system should be able to roll back or has a persistent cache.Firstly, a SSD cache FPC, is designed to avoid accidental cache data loss. PFC uses modified set-associative address mapping to fully explore hard disk’s performance. PFC also use non-volatile memory(SSD) and update in-SSD metadata to ensures the cache data consistency when the system restarts or crash.Secondly, we proposed a new metadata update method, LUFUW. LUFUW is inspired by confirmation page piggyback strategy in TCP protocol, which is widely used in internet technology. In the protocol, the metadata updates when a new operation is bringing back to SSD metadata area. In this way, most of the blocks are available after system restart. Its performance is2times faster than Flashcache after system restart.The last design is ST-CDP, which has the ability to trace back and to continually protect data. Traditional continuous data protection generates large amount of record to save every change in the system. The new ST-CDP brings in XOR to improve data compression ratio for the sake of saving storage space. Meanwhile, it combines snapshot technology and current continuous data protection method, which Inserts snapshot into parity log to record data change. In this way, the system can rapidly recover reliable data. This paper also presents the mathematic model about when and how to insert snapshot, optimize the storage space and recovery time. Based on the mathematic model, we design and implement a continuous data protection in linux cornel.
Keywords/Search Tags:Continuous Data Protection, Cache, Metadata Management, Hot Restart, SolidState Drive, I/O Performance
PDF Full Text Request
Related items