File and Storage Technologies · 2005 · 28 citations · 8 references
ComputingComputer ScienceComputer SystemsFault ToleranceFault-tolerant SystemsOperating SystemsData RecoveryData RedundancyFault-tolerant Storage SystemsReliable System DesignSoftware Fault ToleranceStorage ConsistencyCrash AnalysisCrash ConsistencyJournalingJournaling File SystemsRaid ReliabilityRaid SystemsRedundant ComputationResynchronization
We investigate the problem of slow, scan-based, software RAID resynchronization that restores consistency after a system crash. Instead of augmenting the RAID layer to quicken the process, we leverage the functionality present in a journaling file system. We analyze Linux ext3 and introduce a new mode of operation, declared mode, that guarantees to provide a record of all outstanding writes in case of a crash. To utilize this information, we augment the software RAID interface with a verify read request, which repairs the redundant information for a block. The combination of these features allows us to provide fast, journal-guided resynchronization. We evaluate the effect of journal-guided resynchronization and find that it provides improved software RAID reliability and availability after a crash, while suffering little performance loss during normal operation.
8