Electrical Engineering
      and Computer Sciences

Electrical Engineering and Computer Sciences

COLLEGE OF ENGINEERING

UC Berkeley

Performance and Reliability in Redundant Arrays of Inexpensive Disks

Garth A. Gibson

EECS Department
University of California, Berkeley
Technical Report No. UCB/CSD-89-535
September 1989

Arrays of magnetic disks, or disk arrays, are becoming more common for a variety of reasons. As overall system processing speeds grow at a rate far ahead of magnetic disk technology, disk arrays can provide much needed increases in data throughput (megabytes per second) and access throughput (accesses per second). Disk arrays capitalize on the smaller form-factor magnetic disks that are rapidly becoming denser and cheaper per megabyte than conventional IBM 3380 class disks. Moreover, a single high quality small disk design can be used as the basis of a range of price and performance disk products when arrays are employed. Disk arrays also offer the opportunity to achieve high data reliability at a lower cost than conventional methods based on complete duplication of data. After an extended review of disk array motivations, organizations, and performance expectations, this paper examines the data reliability of a Redundant Array of Inexpensive Disks (RAID) in comparison to conventional disks. In particular, we include support hardware failures (power, cooling, controllers and cabling), spare pool depletion, and non-exponential disk lifetimes. This paper finds that data redundancy deals with disk failures quite effectively, leaving support hardware failures as the reliability bottleneck; that a relatively small spare pool works quite well; and that although reliability is sensitive to disk lifetime distributions, the effect is only pronounced for small spare pools with slow refill rates.


BibTeX citation:

@techreport{Gibson:CSD-89-535,
    Author = {Gibson, Garth A.},
    Title = {Performance and Reliability in Redundant Arrays of Inexpensive Disks},
    Institution = {EECS Department, University of California, Berkeley},
    Year = {1989},
    Month = {Sep},
    URL = {http://www.eecs.berkeley.edu/Pubs/TechRpts/1989/5910.html},
    Number = {UCB/CSD-89-535},
    Abstract = {Arrays of magnetic disks, or disk arrays, are becoming more common for a variety of reasons. As overall system processing speeds grow at a rate far ahead of magnetic disk technology, disk arrays can provide much needed increases in data throughput (megabytes per second) and access throughput (accesses per second). Disk arrays capitalize on the smaller form-factor magnetic disks that are rapidly becoming denser and cheaper per megabyte than conventional IBM 3380 class disks. Moreover, a single high quality small disk design can be used as the basis of a range of price and performance disk products when arrays are employed. Disk arrays also offer the opportunity to achieve high data reliability at a lower cost than conventional methods based on complete duplication of data. After an extended review of disk array motivations, organizations, and performance expectations, this paper examines the data reliability of a Redundant Array of Inexpensive Disks (RAID) in comparison to conventional disks. In particular, we include support hardware failures (power, cooling, controllers and cabling), spare pool depletion, and non-exponential disk lifetimes. This paper finds that data redundancy deals with disk failures quite effectively, leaving support hardware failures as the reliability bottleneck; that a relatively small spare pool works quite well; and that although reliability is sensitive to disk lifetime distributions, the effect is only pronounced for small spare pools with slow refill rates.}
}

EndNote citation:

%0 Report
%A Gibson, Garth A.
%T Performance and Reliability in Redundant Arrays of Inexpensive Disks
%I EECS Department, University of California, Berkeley
%D 1989
%@ UCB/CSD-89-535
%U http://www.eecs.berkeley.edu/Pubs/TechRpts/1989/5910.html
%F Gibson:CSD-89-535