DocumentCode :
723698
Title :
D-Code: An Efficient RAID-6 Code to Optimize I/O Loads and Read Performance
Author :
Yingxun Fu ; Jiwu Shu
Author_Institution :
Dept. of Comput. Sci. & Technol., Tsinghua Univ., Beijing, China
fYear :
2015
fDate :
25-29 May 2015
Firstpage :
603
Lastpage :
612
Abstract :
With the reliability requirement increasingly important, RAID-6, which can tolerate any two concurrent disk failures, has been widely used in various storage systems. One class of typical RAID-6 implementations is to use Maximum Distance Separable (MDS) erasure codes. However, most existing RAID-6 MDS codes suffer from unbalanced I/O or high I/O cost, and cannot provide satisfied read performance on both normal mode and degraded mode. All these metrics are important in modern storage systems. In this paper, we propose a new RAID-6 MDS code termed D-Code to address these problems. D-Code uses a new kind of horizontal parities to increase the possibility of continuous data elements sharing the common parities in order to provide low I/O cost and good degraded read performance, while uses deployment parities to assure all parities can be evenly distributed in the last two rows in order to achieve good load balancing and good normal read performance. Our evaluations and experiments show that D-Code not only provides good load balancing and low I/O cost under different workloads, but also achieves good performance on both normal reads and degraded reads. E.g., D-Code achieves up to 21.3% and 13.5% higher read speed than RDP code and H-Code in normal mode, while gains up to 26.0% higher read speed than X-Code in degraded mode.
Keywords :
input-output programs; resource allocation; storage management; D-Code; I/O load performance optimization; I/O read performance optimization; MDS erasure codes; RAID-6 MDS codes; concurrent disk failures; deployment parities; horizontal parities; load balancing; maximum distance separable erasure codes; storage systems; Encoding; Fault tolerance; Fault tolerant systems; Load management; Mathematical model; Nickel; Switches; Degraded Read; Erasure Code; I/O Cost; Load Balancing; Normal Read; RAID-6;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Parallel and Distributed Processing Symposium (IPDPS), 2015 IEEE International
Conference_Location :
Hyderabad
ISSN :
1530-2075
Type :
conf
DOI :
10.1109/IPDPS.2015.46
Filename :
7161548
Link To Document :
بازگشت