March 26, 2011 at 1:05 pm
happycat59 (3/25/2011)
I believe that the data is actually stored on 3 disks...2 have the actual data and the 3rd has the exclusive OR of the other 2 disks. Using this, if you loose any one of the 3 disks that store the data you are after, the other 2 have enough information to give to retrieve the data. This basic pattern is used regardless of the number of disks in the array. When there are more than 3 disks, each chunk of data is still stored on 3 of the disks. The controller manages the allocation of disk space to ensure that all disks on the array are utilised.
I can tell you for sure that using EMC technology and I believe it was HP configuration manager (don't quote me on that, it's been about 6 years since I worked at this specific company) that the RAID 5 does NOT work in chunks of 3 but stripes across all 14 of the spindles if you so choose. Also, one drive is not the dedicated parity, but it definately chunked it out across the array, a little here, a little there.
Now, others might do it differently and a single parity drive makes more sense (to me) then playing whack-a-mole with the parity locations, but I didn't get to design it. 🙂
... The amount of research you end up doing while watching two transactional replications rebuild their snapshot on a 2 tera system because you lost two drives in the RAID 5 to decide if you should shoot your networking team is amazing...
Never stop learning, even if it hurts. Ego bruises are practically mandatory as you learn unless you've never risked enough to make a mistake.
For better assistance in answering your questions[/url] | Forum Netiquette
For index/tuning help, follow these directions.[/url] |Tally Tables[/url]
Twitter: @AnyWayDBA
March 26, 2011 at 1:07 pm
UMG Developer (3/25/2011)
happycat59 (3/25/2011)
Steve - whilst you may not have heard of checksum (Vs parity) in relation to RAID 5 , try searching google for "RAID 5 checksum". I do agree that you can view this as a "parity" but I am one of them dinosaurs who have been around since the days of punch cards, paper tape and magnetic tape. Parity back then was more to do with detecting error.Isn't that still what the parity/checksum in RAID5 is for to detect and correct errors?
No, it's never used for detection, only for correction.
Incidentally, I suspect that I'm a much older dinosaur that happycat58 and for me "parity" has always meant "an extra bit that tells you whether the rest of the bits have an even number or odd number of 1s" which is exactly what parity on a RAID5 array is; you could use that for detecting error, yes, and on paper tape and cards longitudinal parity was often used for that (and was close to useless, since common hardware problems tended to produce error patterns that that wouldn't detect); but of course you couldn't use horizontal parity (which would have been rather less useless) if the number of bits on your horizontal line was the number of data bits you needed, which it almost always was for binary data.
On a single unit medium like cards or tape horizontal parity (which I'm sure is what happycat59 was thinking of, there was a reference to byte parity somewhere) was useless for error correction, because if you detected an error it could be that the parity bit or any of the data bits was wrong; in RAID5 horizontal parity is fine for error correction, because the error detected is that a drive has packed up - so we know which bits are OK and which or wrong. And if you have 9 drives, you can think of RAID5 parity as being just like byte parity on a 9 channel tape - exceptthat (a) which channel is the parity swaps at regular intervals and (b) when you do recovery you know which of the bits in each row is the failed one.
BTW, don't take my explaining RAID5 as a sensible use of parity means I approve of the thing - if data reliability and integrity mean more to you than a couple of per cent on your hardware budget eschew RAID 5 and use RAID 10.
Tom
March 26, 2011 at 1:20 pm
happycat59 (3/25/2011)
I believe that the data is actually stored on 3 disks...2 have the actual data and the 3rd has the exclusive OR of the other 2 disks. Using this, if you loose any one of the 3 disks that store the data you are after, the other 2 have enough information to give to retrieve the data. This basic pattern is used regardless of the number of disks in the array. When there are more than 3 disks, each chunk of data is still stored on 3 of the disks. The controller manages the allocation of disk space to ensure that all disks on the array are utilised.
No chunk of data is ever stored on 3 discs. In a 3 disc array each bit is stored on one disc, and the XOR of each even numbered bit and the next following bit is stored on whichever disc neither of those bits was stored on.
I think anyone guilty of describing your remarkable multi-disc arrangement as with 3-way parity no matter how many drives are used either is thinking of RAID 05 (RAID 0+5) (a really imbecile arrangement, dreamt up by a gentleman who didn't understand what RAID 50 was) or isn't thinking at all.
Tom
Viewing 3 posts - 16 through 17 (of 17 total)
You must be logged in to reply to this topic. Login to reply