臺灣博碩士論文加值系統

As the process progresses, SSD costs continue to reduction. From the MLC to TLC or even the future launch of the QLC nand flash memory chip is gradually replacing the traditional hard drive. However, increasing the density of the chip to achieve high capacity must be affected by reliability degradation. RAID technology is designed to enhance the reliability and performance of storage systems. Unfortunately, there are some drawbacks to applying a RAID structure to an SSD. RAID technology balances requests for each storage device to improve its performance. In addition, the FTL implementation of the wear balance mechanism will cause the SSDs in the RAID array to age at the same rate, and cause the device to fail at the same time. If the faulty device exceeds the RAID tolerance range, it will lead to data loss occurs. In addition, RAID uses the parity protection mechanism, which means that when the corresponding user data is updated, the parity must be updated at the same time. When the user data is hot, too much overhead will lead to performance degradation. In this paper, we propose a novel SSD-based RAID storage system based on a system-level wear-unleveling mechanism to prevent data loss. And the use of the concept of spare devices to delay the aging rate of partial devices, thus preventing the devices failure at the same time. We also propose a difference from the traditional RAID write mechanism, which breaks the rules that must be updated with user data to update the parity. So the parity is no longer frequent updates, performance will be improved. The experimental results show that the proposed mechanism can maintain the probability of data loss at 5.1e-3. Although compared with the Diff-RAID slightly down, but the performance is still increased by 43.7% -63.6%.

As the process progresses, SSD costs continue to reduction. From the MLC to TLC or even the future launch of the QLC nand flash memory chip is gradually replacing the traditional hard drive. However, increasing the density of the chip to achieve high capacity must be affected by reliability degradation. RAID technology is designed to enhance the reliability and performance of storage systems. Unfortunately, there are some drawbacks to applying a RAID structure to an SSD. RAID technology balances requests for each storage device to improve its performance. In addition, the FTL implementation of the wear balance mechanism will cause the SSDs in the RAID array to age at the same rate, and cause the device to fail at the same time. If the faulty device exceeds the RAID tolerance range, it will lead to data loss occurs. In addition, RAID uses the parity protection mechanism, which means that when the corresponding user data is updated, the parity must be updated at the same time. When the user data is hot, too much overhead will lead to performance degradation. In this paper, we propose a novel SSD-based RAID storage system based on a system-level wear-unleveling mechanism to prevent data loss. And the use of the concept of spare devices to delay the aging rate of partial devices, thus preventing the devices failure at the same time. We also propose a difference from the traditional RAID write mechanism, which breaks the rules that must be updated with user data to update the parity. So the parity is no longer frequent updates, performance will be improved. The experimental results show that the proposed mechanism can maintain the probability of data loss at 5.1e-3. Although compared with the Diff-RAID slightly down, but the performance is still increased by 43.7% -63.6%.

0.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
0.2 Background and Related Work . . . . . . . . . . . . . . . . . . . . 8
0.2.1 Redundant Array of Independent Disks 5 . . . . . . . . . . 8
0.2.2 Dierential RAID . . . . . . . . . . . . . . . . . . . . . . . 8
0.2.3 Flash Aware RAID . . . . . . . . . . . . . . . . . . . . . . 8
0.3 Locality-Aware RAID . . . . . . . . . . . . . . . . . . . . . . . . . 9
0.3.1 Overview of System Architecture . . . . . . . . . . . . . . 9
0.3.2 Stripe Management . . . . . . . . . . . . . . . . . . . . . . 10
0.4 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
0.4.1 Experiment Setup . . . . . . . . . . . . . . . . . . . . . . . 17
0.4.2 Bit Error Rate Module . . . . . . . . . . . . . . . . . . . . 19
0.4.3 Average Write Response Time . . . . . . . . . . . . . . . . 19
0.4.4 Average Read Response Time . . . . . . . . . . . . . . . . 20
0.4.5 Total Number of Write Operations . . . . . . . . . . . . . 20
0.4.6 Performance with Dierent Cache Size . . . . . . . . . . . 20
0.4.7 Reliability Analysis . . . . . . . . . . . . . . . . . . . . . . 21
0.4.8 Reliability Analysis with Dierent Parameter . . . . . . . 22
0.4.9 Reliability Analysis with Dierent Spare Device . . . . . . 22
0.4.10 Reliability Analysis with Dierent Initial Device . . . . . . 23
0.4.11 Average Life Page Copy Count of Each Erase . . . . . . . 23
0.4.12 Device Overhead . . . . . . . . . . . . . . . . . . . . . . . 23
0.4.13 Reconstruction Time . . . . . . . . . . . . . . . . . . . . . 24
0.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

[1] M. Balakrishnan, A. Kadav, V. Prabhakaran, and D. Malkhi, \Dieren-
tial raid: Rethinking raid for ssd reliability," ACM Transactions on Storage
(TOS), 2010.
[2] S. Im and D. Shin, \Flash-aware raid techniques for dependable and high-
performance
ash memory ssd," IEEE Transactions on Computers, vol. 60,
no. 1, pp. 80{92, Jan 2011.
[3] Hsieh, Jen-Wei, and M.-X. Liu, \Congurable reliability framework for
ssd-raid," Non-Volatile Memory Systems and Applications Symposium
(NVMSA), 2014.
[4] J. Kim, J. Lee, and J. Choi, \Improving ssd reliability with raid via elastic
striping and anywhere parity," Dependable Systems and Networks (DSN),
2013.
[5] Y. Pan, Y. Li, and Y. Xu, \Grouping-based elastic striping with hotness
awareness for improving ssd raid performance," Dependable Systems and Net-
works (DSN), 2015.
[6] \Micronl84128gb,256gb,512gbasync/syncenterprisenandfeatures." [Online].
Available: http://www.micron.com/parts/nand-
ash/enterprise-nand/
[7] UMassTraceRepository, \Oltp applications of two nancial institutions."
[Online]. Available: http://traces.cs.umass.edu/index.php/Storage/Storage
[8] L.-P. Chang and C.-D. Du, \Design and implementation of an ecient wear-
leveling algorithm for solid-state-disk microcontrollers," ACM Transactions
on Design Automation of Electronic Systems (TODAES), 2009.
[9] L.-P. Chang, \A hybrid approach to nand-
ash-based solid-state disks,"
IEEE Transactions on Computers, 2010.
[10] J.-W. Hsieh, C.-W. Chen, and H.-Y. Lin, \Adaptive ecc scheme for hybrid
ssds," IEEE Transactions on Computers, 2015.
[11] L. M. Grupp, A. M. Cauleld, and J. Coburn, \Characterizing
ash mem-
ory: Anomalies, observations, and applications," IEEE/ACM International
Symposium on, 2009.