On Thu, Jun 05, 2008 at 08:16:26AM -0400, David Lee Lambert wrote:
I disagree with the statement that "RAID technology [...] does not detect if a specific bit on a SATA drive becomes unreadable." If a bit is unreadable or reads back as the wrong value, RAID-5 (or NetApp's RAID-DP) can detect and fix the error when the data is read back.
*bzzt* nope. RAID by itself does not "detect" bad data. It can correct bad data when it's detected (by other means, usually the hardware driver), but it doesn't detect it.
At least, that's not how it's implemented in all RAID systems that I'm aware of. To _detect_ bad data with RAID, you'd have to read an entire track, and verify that the checksum is correct. If it isn't, then AT LEAST ONE of the sectors is in error, but it's impossible to determine which one, without additional information.
RAID-DP might distill that info from the diagonal checksums (if it's a single sector error), but you can hardly expect your super fast networked storage hardware device to go and solve crossword puzzles for you every time you request a block of data.
RAID works because drives don't just flip bits, they fail. Or the CRC on the drive block fails. In any way, there's some indication that something is amiss on a certain drive. RAID then offers the ability to replace the failed data from the other drives, using the parity.
(Note that if the data is never read back, then it's immaterial whether it's correct.)
Deduplication does not increase this risk. In fact, deduplication means that the duplicated data is read back more often, which should mean that any errors that occur would be detected sooner.
On Thursday 05 June 2008 06:54:15 am Chong, Jenson wrote:
Hi,
Anybody have insights on this article? Does our ONTAP "Lost Write Protection" address this issue?
Regards, Jenson
A bit of a flaw with SATA disk drives June 03, 2008 By Jerome Wendt Network World Asia High-capacity serial ATA (SATA) disk drives are now amainstay in many storage systems and make it feasible for almost any company to obtain a storage system with terabytes of capacity at a reasonable cost. Yet these systems reveal a specific, known deficiency of SATA disk drives that demand companies exercise caution as to what environments they deploy these systems into.
A minor flaw with SATA disk drives that high capacitystorage systems expose is their bit error rate. Bit errors occur infrequently - about once for every 100 trillion bits. However RAID technology, which is normally used by storage systems to protect against data loss, does not detect if a specific bit on a SATA drive becomes unreadable.
While this is normally not a problem on smaller systems,as storage systems add more capacity, the issue becomes more acute. On systems with more than 10TB of capacity the probability of a specific bit of data becoming unreadable is a distinct possibility. On systems with over 100TB, it becomes almost a certainty.
So the question becomes: Does losing access to one bitof data really matter? Often, it doesn't unless one stores deduplicated data on these systems which is now the fastest growing trend in data storage. When data is deduplicated, the storage system's need to read every bit of data becomes paramount. The inability to access even a bit of data can result in multiple files becoming unreadable since they all may depend on a specific bit of data to complete their reconstruction.
High capacity SATA-based storage systems are the answerto many companies' archiving and backup problems. But SATA bits can bite and using SATA drives to store large amounts of deduplicated data is not always the match made in heaven that vendors make them out to be.
Jerome Wendt is the president and lead analyst at DCIGInc. You may read his blogs at www.dciginc.com http://www.dciginc.com/ .
--
David L. Lambert Software Developer, Precision Motor Transport Group, LLC Work phone 517-349-3011 x215 Cell phone 586-873-8813