Research Of Reliability Reinforcements For SSD And RAID

In order to increase the storage density,the process size of Flash memory is decreasing.However,the probability of bit errors occurred in Flash memory is increasing and error correction codes(ECCs)in SSD are reluctant to protect the storage data.This paper proposes reinforcements in bit-level,sector-level and drive-level against the bit errors,latent sector errors(LSEs)in SSD and the drive aging problem in All Flash Array(AFA),respectively.The main work is as follows.Chapter 2 proposes a measurement to evaluate the reliability of a single SSD based on data loss probability and a method to estimate its residual lifetime.As the reliability of Flash memory is declining with abrasion,the dependability of SSD is decreasing correspondingly.The chapter carries out tests on NAND Flah chips(NANDs)in detail and finds that the program/erase(P/E)cycles and retention time are main factors which dominate the page raw bit error rate(RBER).To analyse the reliability of SSD,we establish a novel bit error model for NANDs.It calculates data loss probability and residual lifetime according to the distribution and growth model of Flash page RBER,respectively.They are able to precisely forecast the reliability of a single SSD.Chapter 3 proposes a Flash page life-aware scrubbing scheme for SSD.The chapter first gives a detailed analysis on the cumulative growth law of retention error.Combined with the ECCs in SSD,the bit errors in Flash pages are scrubbed periodically.In order to reduce the polling overhead,the proposed method is only concerned about the risky pages which are more likely to lose data.Then,different schemes are devised to scrub retention errors.For hot blocks,a shortened ECC scheme is adopted to improve the correctability of ECCs in SSD.For cold blocks,a bit error log method is designed to preserve the bit error locations and flip directions.Simulation results show that the proposed scheme is able to control the risk of vulnerable pages and significantly enhance the reliability of SSD at very small overhead in time.Chapter 4 proposes a Flash-aware Intra-disk Redundancy(FAIDR)scheme based on adaptive Reed-Solomon(RS)codes.The latent sector errors and drive failures are major errors in AFA.The chapter combines the FAIDR with RAID technology to combat them simultaneously.For saving parity space,we devise an adaptive RS code in which the ratio of redundancy grows with the aging of NANDs.In order to eliminate the parity handling overhead,we establish all FAIDR stripes on the physical page number of blocks regardless of the page state.Simulation results show that the proposed scheme apparently improves the AFA's reliability and prolongs its lifetime at small performance and space overhead.Chpater 5 proposes a new differential RAID(Diff-RAID)scheme to enhance the reliability of AFA.The synchronously aging problem of SSD brings about great risk to the AFA.In the proposed scheme,we use the capacity differential to form a age differential.To stabilize the reliability of AFA,the worn drives are continually replaced by new drives.The young drives protect the aging drives by using RAID technology.Compared with parity-based Diff-RAID,the age differential in our scheme is independent of workload,the time-consuming of reconstruction is greatly reduced and the I/O performance is better.In this paper,we detailed analyze the error characteristics of NANDs.Combined with the special operating manner of NANDs and structure of SSD,we propose different levels of reinforcements to cope with the increasing bit errors and LSEs in SSD and the aging problem of SSD at very small performance and space overhead.