Buying a MAID to take care of your storage

Remember RAID? Of course you do. But I'll bet you're thinking of the RAID that stands for redundant arrays of independent disks. I'm thinking of the RAID that stands for redundant arrays of inexpensive disks -- the original RAID.

The original was born on a university campus by a brilliant group of hackers (I use the term with all due respect) who wanted to see if they could turn a pile of disks they had lying around the computing lab into one big disk. The rest is storage history.

RAID was morphed from inexpensive to independent by marketing folks who believed IT buyers would equate "inexpensive" with "cheap and unreliable." The original RAID had its early detractors, who said it wouldn't work for production IT. That's how it is with technological innovations that:

  1. Appear too simple and obvious to be real
  2. Threaten to significantly undermine an existing and well-entrenched monetization model

Hold those thoughts for a moment while I introduce you to RAID's most recent grandchild -- MAID, for massive array of idle disks.

Like the original RAID, MAID was born in an academic setting by another group of brilliant hackers that also included a university professor. MAID reduces disk power-on hours by keeping the majority of data disks within an array powered off (idle) at any given time. When the data on an idle disk is required by an application, the disk is spun up to service the request. That's a big deal, considering the rising cost of energy, coupled with the need to recover more data for ever longer periods of time and do so in a manner that doesn't require days of searching when the data is needed.

A MAID subsystem made up of Serial Advanced Technology Attached (SATA) disks is therefore a prime candidate for certain backup and very long-term (seven to 10 years) archival storage applications. However, like the original RAID, MAID also has its vocal detractors.

The knock on MAID basically goes like this: The spin-up/spin-down cycles used in MAID arrays make these arrays inherently unreliable. This stems from the assumption that disk failures are more likely to occur during power-down/power-up sequences when the drives have been inactive for extended time periods -- problems caused by head-disk stiction and/or corrosion. Granted. But, MAID engineers have done things to overcome this problem like periodically spinning up the idle disks to verify data and drive.

The first company to bring MAID to market -- Copan Systems Inc. -- patented a number of MAID-related processes to make MAID a contender in a production IT setting. I don't have room in this column to describe them. What I can report is that they work. I say this because Copan is about to release a detailed and, I believe, statistically valid study of MAID/SATA disk reliability in the field. To summarize its conclusions:

  1. Copan Systems reports that its annualized MAID/SATA drive-failure rate has been less than 0.4% over the 20-month period under study.
  2. This failure rate equates to an effective mean time between failures (MTBF) of 2.3 million hours.
  3. Presently, this failure rate is approximately one-fourth that of many SATA manufacturers' specified nominal MTBF ratings for an always-on SATA drive.

The study was needed to validate MAID in the same way that RAID was validated -- in production IT settings. I think we'll now be seeing more MAIDs.

John Webster is senior analyst and founder of research firm Data Mobility Group LLC. He is also the author of numerous articles and white papers on a wide range of topics and is the co-author of the book Inescapable Data: Harnessing the Power of Convergence (IBM Press, 2005). Webster can be reached at jwebster@datamobilitygroup.com.

Related:

Copyright © 2006 IDG Communications, Inc.

  
Shop Tech Products at Amazon