USENIX 2003 Annual Technical Conference, FREENIX Track Abstract
Pp. 151-164 of the Proceedings |
StarFish: Highly-Available Block Storage
Awarded Best Paper!
Eran Gabber, Jeff Fellin, Michael Flaster, Fengrui Gu, Bruce Hillyer, Wee Teck Ng, Banu Özden, and Elizabeth Shriver, Lucent Technologies Bell Laboratories
Abstract
In this paper we present StarFish, a highly-available geographically-dispersed block storage system built from commodity servers running FreeBSD, which are connected by standard high-speed IP networking gear. StarFish achieves high availability by transparently replicating data over multiple storage sites. StarFish is accessed via a host-site appliance that masquerades as a host-attached storage device, hence it requires no special hardware or software in the host computer. We show that a StarFish system with 3 replicas and a write quorum size of 2 is a good choice, based on a formal analysis of data availability and reliability: 3 replicas with individual availability of 99%, a write quorum of 2, and read-only consistency gives better than 99.9999% data availability. Although StarFish increases the per-request latency relative to a direct-attached RAID, we show how to design a highly-available StarFish configuration that provides most of the performance of a direct-attached RAID on an I/O-intensive benchmark, even during the recovery of a failed replica. Moreover, the third replica may be connected by a link with long delays and limited bandwidth, which alleviates the necessity of dedicated communication links to all replicas.
- View the full text of this paper in HTML or
PDF.
The Proceedings are published as a collective work, © 2003 by the USENIX Association. All Rights Reserved. Rights to individual papers remain with the author or the author's employer. Permission is granted for the noncommercial reproduction of the complete work for educational or research purposes. USENIX acknowledges all trademarks within this paper.
- If you need the latest Adobe Acrobat Reader, you can download it from Adobe's site.
|