容錯式磁碟陣列5普遍被使用於企業級的儲存系統，藉由儲存額外的校驗資訊以提昇資料的校正能力，並且在成本與效能之間取得一個平衡。然而當固態硬碟逐漸取代傳統硬碟的時代，採用固態硬碟的RAID-5儲存系統仍未廣泛的普及於市場。其原因在於當 RAID-5 更新資料時，若更新資料的範圍小於條帶集，需要讀出舊有的資料與校驗資訊，才能重新計算校驗資訊，最後寫入新資料與校驗資訊。由於SSD的out-place-update的特性，頻繁的更新校驗資訊將衝擊固態硬碟有限的使用壽命。另外，讀取舊資料與頻繁更新校驗資訊也導致整體系統效能下降。

本文提出一個局部校驗資訊緩衝管理機制(Parity Buffer Management Scheme for Hybrid RAID Storage Systems，PBM-RAID)，藉由RAID控制器常用的非揮發性記憶體作為局部校驗資訊的快取記憶體，如此可以解決RAID-5更新資料時，因讀取舊資料以及頻繁更新校驗資訊等因素，導致整體系統效能下降的問題；另外，當更新資料分布於不同條帶集時，由於每一條帶集皆需要利用PPC存放局部校驗資訊，如此PPC的空間很快就被用完，需要將較久未被更新的局部校驗資訊回存於SSD之中。如此將導致PPC的命中率降低，無法發揮延遲寫入校驗資訊的特點。因此，我們利用HDD低成本與大容量的特性來作為校驗資訊主要的緩衝區，以解決 PPC 在頻繁更新時容易遇到空間不足的問題。並且藉由SSD與HDD平行運作的方式來降低寫入的間接成本。當系統閒置時，則將校驗資訊由快取記憶體轉存至HDD中，或是由HDD回存於SSD中。如此可減少對SSD的寫入成本，並進一步的提升整體寫入效能。實驗結果顯示PBM-RAID將比傳統作法提昇達 51.7%的效能，並且延長SSD壽命達47.1%。

RAID-5 is the most popular RAID technology in use today. In this architecture, one disk is reserved to store the parity, and thus, the multibit burst error in a device can be easily corrected. For this years, solid-state disks(SSDs), which are composed of multiple NAND flash chips, are replacing hard disk drives(HDDs) in the mass storage market. In order to
implement RAID-5 technology in flash storage, we should consider the characteristics of flash memory. However, the parity handling overhead of RAID-5 significantly deteriorates the performance especially in flash memory.

In this paper, we proposed a novel partial parity buffer management scheme and access strategy for hybrid storage systems. To reduce the number of write operations for the parity updates, the proposed scheme delays the parity update which must accompany each data write in the original RAID technique. In addition, the proposed scheme uses the partial parity technique to reduce the number of read operations required to calculate
a parity. We also flush the partial parity in a buffer ,which is composed of several HDDs, to release the space of cache for future updates. We provide evaluation results based on a trace-driven simulator. The proposed scheme improved write performance up to 51.7% and 47.8%, compare with RAID-5 and FA-RAID respectively. And it also reduced the overall erase count up to 47.1% and 24.1%, compare with RAID-5 and FA-RAID respectively.

[1] “Micron l84 128gb, 256gb, 512gb async/sync enterprise nand features,” http://www.micron.com/parts/nand-flash/enterprise-nand/.
[2] “Seagate st500dm002 500gb hdd datasheet,” http://www.seagate.com/
www-content/product-content/barracuda-fam/desktop-hdd/barracuda-7200-12/
zh-tw/docs/desktop-hdd-ds1770-4-1405tw.pdf.
[3] “Ddr3 sdram wiki,” http://en.wikipedia.org/wiki/DDR3_SDRAM.
[4] J.-W. Hsieh, H.-Y. Lin, and D.-L. Yang, “Multi-channel architecture-based ftl for reliable and high-performance ssd,” IEEE Transactions on Computers, vol. 99, no. PrePrints, p. 1, 2013.
[5] U. T. Repository, “Oltp applications of two financial institutions.” [Online]. Available: http://traces.cs.umass.edu/
[6] W.-H. Lin and L.-P. Chang, “Dual greedy: Adaptive garbage collection for page-mapping solid-state disks,” in Design, Automation Test in Europe Conference Exhibition (DATE), 2012, March 2012, pp. 117–122.
[7] S. Zertal, “A reliability enhancing mechanism for a large flash embedded satellite storage system,” in Systems, 2008. ICONS 08. Third International Conference on, April 2008, pp. 345–350.
[8] S. Im and D. Shin, “Flash-aware raid techniques for dependable and high-
performance flash memory ssd,” Computers, IEEE Transactions on, vol. 60, no. 1, pp. 80–92, Jan 2011.
[9] J. Guo, W. Wen, Y. Z. Li, S. Li, H. Li, and Y. Chen, “Da-raid-5: A disturb aware data protection technique for nand flash storage systems,” in Design, Automation Test in Europe Conference Exhibition (DATE), 2013, March 2013, pp. 380–385.
[10] B. Mao, H. Jiang, D. Feng, S. Wu, J. Chen, L. Zeng, and L. Tian, “Hpda: A hybrid parity-based disk array for enhanced performance and reliability,” in Parallel Distributed Processing (IPDPS), 2010 IEEE International Symposium on, April 2010, pp. 1–12.
[11] “FUJITSU RAID introduction,” http://www.fujitsu.com/global/products/
computing/storage/eternus/glossary/raid/raid5.html, 2013.
[12] M.-X. L. Jen-Wei Hsieh, “Configurable reliability framework for ssd-raid,” in the 3rd IEEE Nonvolatile Memory Systems and Applications Symposium (NVMSA 2014), August 2014, pp. 20–21.
[13] “All SSD Charts 2013,” http://www.tomshardware.com/charts/ssd-charts-2013/benchmarks,129.html, 2013, [Online; accessed 19-Aug-2014].
[14] “All Enterprise HDD Charts,” http://www.tomshardware.com/charts/
enterprise-hdd-charts/benchmarks,156.html, 2013, [Online; accessed 19-Aug-
2014].
[15] J. Hruska, “Storage Pricewatch: Get out your wallet, hard drive and
SSD prices have dropped,” http://www.extremetech.com/computing/164677-storage-pricewatch-hard-drive-and-ssd-prices-drop-making-for-a-good-time-to-buy, 2013, [Online; accessed 19-Aug-2014].
[16] B. Mao, D. Feng, S. Wu, J. Chen, L. Zeng, and L. Tian, “Raid10l: A high performance raid10 storage architecture based on logging technique,” in Computer Systems Architecture Conference, 2008. ACSAC 2008. 13th Asia Pacific, Aug 2008, pp. 1–8.
[17] M. Balakrishnan, A. Kadav, V. Prabhakaran, and D. Malkhi, “Differential raid: Rethinking raid for ssd reliability,” Trans. Storage, vol. 6, no. 2, pp. 4:1–4:22, Jul. 2010. [Online]. Available: http://0-doi.acm.org.millennium.lib.ntust.edu.tw/10.1145/1807060.1807061
[18] “Storage Class Memory-IBM Research,” http://researcher.watson.ibm.com/
researcher/view_group.php?id=3631, 2013.
[19] “FRAM vs. MRAM,” http://deltas.blog.com/2013/04/10/fram-vs-mram/, 2013.