Low-Complexity Codes for Random and Clustered High-Order Failures in Storage Arrays
- Creators
-
Cassuto, Yuval
-
Bruck, Jehoshua
Abstract
RC (Random/Clustered) codes are a new efficient array-code family for recovering from 4-erasures. RC codes correct most 4-erasures, and essentially all 4-erasures that are clustered. Clustered erasures are introduced as a new erasure model for storage arrays. This model draws its motivation from correlated device failures, that are caused by physical proximity of devices, or by age proximity of endurance-limited solid-state drives. The reliability of storage arrays that employ RC codes is analyzed and compared to known codes. The new RC code is significantly more efficient, in all practical implementation factors, than the best known 4-erasure correcting MDS code. These factors include: small-write update-complexity, full-device update-complexity, decoding complexity and number of supported devices in the array.
Additional Information
This work was supported in part by the Caltech Lee Center for Advanced Networking.Files
Name | Size | Download all |
---|---|---|
md5:d3ec19045e266c61c1fa32056e8cb047
|
234.0 kB | Preview Download |
md5:16fb510aacd5c8fc538e469e255df70b
|
216.6 kB | Preview Download |
Additional details
- Eprint ID
- 26129
- Resolver ID
- CaltechPARADISE:2009.ETR098
- Created
-
2009-12-04Created from EPrint's datestamp field
- Updated
-
2020-03-09Created from EPrint's last_modified field
- Caltech groups
- Parallel and Distributed Systems Group