Write optimization for RAID5 in sequential data storage

Zhizhuo Sun, Quanxin Zhang*, Yuanzhang Li, Yu'an Tan, Jingyu Liu, Zhongmei Ma

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

3 Citations (Scopus)

Abstract

In sequential data storage, such as video surveillance, continuous data protection (CDP), virtual tape library (VTL), etc., the address of the I/O requests are mainly continuous except a small amount of random accesses, but the small write still exists and hampers the full exploitation of the performance of RAID5. In this paper, a write optimization method for RAID5 in sequential storage (WOSS) is presented. Firstly, address translation (AT) based on area mapping is performed, which maps the discontinuous address of write requests to continuous address space, so the continuous write for RAID5 is realized. Secondly, the write requests are buffered and further reorganized into the new ones aligned to the stripes of RAID5. Then they are dispatched to RAID5, thus the full write for RAID5 is achieved. Full write to RAID5 may completely eliminate the extra overhead caused by read modify write or reconstruction write for generating the parity, meanwhile the continuity of write requests further decreases the seek time and improves the throughput of RAID5 significantly. Experiments show that the optimization may improve the write performance prominently, especially when the 80% workload is sequential, and that the data transfer rate approaches about the maximum of RAID5, while the read performance is limitly reduced. The method is also suitable for RAID4 and RAID6 in sequential data storage.

Original languageEnglish
Pages (from-to)1604-1612
Number of pages9
JournalJisuanji Yanjiu yu Fazhan/Computer Research and Development
Volume50
Issue number8
Publication statusPublished - Aug 2013

Keywords

  • Address translation
  • Disk array
  • RAID5
  • Sequential data storage
  • Small write

Fingerprint

Dive into the research topics of 'Write optimization for RAID5 in sequential data storage'. Together they form a unique fingerprint.

Cite this