Berkeley RAID Paper


1987
Berkeley RAID Paper

U.C. Berkeley paper generated industry-wide definitions for Redundant Arrays of Inexpensive Disks (RAID).

Why it's important

While the RAID paper, "A Case for Redundant Arrays of Inexpensive Disks (RAID)", has received wide recognition, its significance to the industry and technology was debated by the members of the Subsystems Working Group and in the end the group decided to not make the paper a significant event. The RAID paper did define the acronym RAID which subsequently was adopted as a term of the industry.

Discussion:

See Array Technology / STC Alpine article for a discussion of a pioneering RAID product

The RAID paper, "A Case for Redundant Arrays of Inexpensive Disks (RAID)" by David A. Patterson, Randy A Katz and Garth A Gibsonwas presented at the June 1988 ACM SIGMOD Conference; a preprint was published asTechnical Report UCB/CSD 87/391, Berkeley CA, December 1987. The paperhas been recognized as a major contribution to the storage industry as in:

"[T]he paper that has had the most impact (research, products, methodology) over the intervening decade." [1998 ACM SIGMOD Test Of Time Award]

"For the development of the Redundant Array of Inexpensive Disks (RAID) systems" [1999 IEEE Reynold Johnson Information Storage Award to Patterson et. al.]

"Between 1989 and 1993, Patterson and Berkeley colleague Randy Katz led the Redundant Arrays of Inexpensive Disks (raid) project, which resulted in vast improvements in disk system speed and reliability." [Computer History Musuem 2007 Fellow Citation to David Patterson]

There is no dispute that the paper defined the name RAID and created a taxonomy of RAID products (in the paper only RAID 1 thru 5); however. there is a question as to whether the paper contributed anything to the growth of the industry and its technologies beyond the name, since:

  • The five technologies named therein were well known in the industry prior to the paper's publication and products had shipped or were under development independent of the paper

  • Most of the early products used the term disk array or variations thereon rather than the term RAID
    • According to INSPEC from the 1987 first publication of the RAID paper until the end of 1991 there were 22 "array" (no RAID) product announcements compared with 5 RAID product announcements. In 1992 both announcements continued with RAID at 17 exceeding "array" at 11.
  • The paper proposes the RAID replacement of SLEDs (Single Large Expensive Disks); however, virtualization of mainframe disks -- a key technology necessary for such replacement is not mentioned in the paper. SLEDs were then primarily an IBM mainframe product which had extensive format and timing requirements (the CKD variable record length format) which could not be met by the RAID arrays proposed. Virtualization of CKD records onto fixed blocks was not invented until several years after the RAID paper. Furthermore, large arrays also require virtualization to present multiple virtual fixed block drives to the system.

  • The industry leaders such as EMC and Auspex did not initially use the term RAID. By 1995 EMC had 40% share of the IBM mainframe disk storage market without once advertising its products as RAID subsystems, preferring instead “Integrated Cached Disk Array” (ICDA)."

  • Most of the early technical publications used used the term disk array or variations thereon rather than the term RAID
    • According to Google Scholar from 1985 to 1991 there were 133 cites to the RAID paper and 364 cites to disk arrays with no mention of RAID, about 90 prior to the RAID paper publication.
That is, it is clear the disk array industry was founded and growing without the RAID name so that its subsequent universal adoption maybe coincidence rather than causation. It is certainly possible that the early financial success of industry leaders such as EMC in the mainframe market and Auspex in the UNIX server market had more to do with the industry's growth than the change in name from disk arrays to RAID.

Even the term RAID turned out to be a misnomer, It soon became obvious that the disk arrays would be Redundant Arrays of Independent Disks, not inexpensive disks, since the then low cost disks were not suitable for enterprise applications and the later added RAID-0 did not have any redundancy.

Additional Information

Patent activity pre-dating RAID paper

Priority datePatent NumberTitlePublication date
June 2, 1987WO 1988009968 A1Fault-tolerant, error-correcting storage systemDecember 15, 1988
June 12, 1986 US 4,761,785Parity Spreading To Enhance Storage AccessAugust 2, 1988
May 31, 1977US 4,092,732 ASystem for recovering data stored in failed memory unitMay 30, 1978

RAID: A Personal Recollection of How Storage Became a System, R. Katz, November 18, 2010, IEEE
Provenance note:
The Subsystem Working Group of the Computer History Museum's Storage SIG discussed the RAID paper in 2009 and at that time voted to not include it on in their Timeline Of Signficicant Events. Consequently a stub article was created on April 10, 2009, and subsequently updated in 2012 to a Version 11 of this article which attempts to reflects the 2009 discussion based upon a review of the 2009 emails and minutes of the WG. A minority position within the WG was that the standardizationof the name set which enabled comparison acrossproduct lines was a significant contributor to the growth of the industry and thereby qualified as a significant event.

More pages