[Sussex] Recovering from RAID 5
D.Morris at brighton.ac.uk
D.Morris at brighton.ac.uk
Tue Nov 9 11:47:00 UTC 2010
Hi Steve,
Steve Dobson wrote:
> Hi Dave
>
> On 09/11/10 11:02, D.Morris at brighton.ac.uk wrote:
>> Steve Dobson wrote:
>>> But is the performance hit you take with calculating two parity blocks
>>> worth it?
>>> In the days of old when disks where small and RAID 5 was useful, because
>>> you only paid 1/nth the cost for increased tolerance (where n is the
>>> number of disks in the set). Then disks where relatively expensive (my
>>> first hard drive cost me around £500 (including the SCSI interface card
>>> for my Amiga 2000) and it was only 46MB unformed.
>>> Thursday I bought 2 500GB disks for less then £100 - because they were
>>> the cheapest disks in the store. I was planing to reinstall my
>>> workstation and I wanted to mirror. That gives me some real protection
>>> for very little write cost. No parity claculation at all, just the need
>>> to write the same block to two different devices.
>>> Mirroring (RAID 1) is a much better solution. What are the changes the
>>> the same block is bad on both disks at the same time? I doubt that it
>>> is that high. RAID 1 works! I had a disk failure in one of my previous
>>> RAID 1 setups (the one with my home partition on has it happens) and I
>>> lost nothing! I just continued using the single disk (my risk) until I
>>> when out and bought a replacement pair.
>>> The problem I find with RAID 1 is that when one disk does fail you can't
>>> get a replacement disk. They just don't do them in that configuration
>>> any more.
>>> Steve
>> I use software RAID 1, which means I just need to get a disk thats bigger, and a make the partition
>> the same size. Actually I think it works if you add a partition which is bigger than the old one.
>> Obviously there a performance cost to this, not sure how much though but I don't notice it on my
>> desktop nor our lightly used development servers at work.
>
> Nice to know that that works. But I feel that when one disk of a RAID
> set goes then the other are likely to be near the end of the reliable
> life too, or close to it. Best to replace the lot anyway.
I tend to use different manufactures to make sure they're not from the same batch.
>
>
>> As for RAID 6 performance loss, it depends upon what the task is for. I use software RAID 6 with 4
>> 1.5TB drives for a NAS with my multimedia stored on it. The machine has a 1000Mbit card, however my
>> desktop which writes to it only has 100Mbit and the network IO of my desktop is the bottleneck. My
>> mythtv frontends can all play back fine whilst I'm writing to it.
>
> That doesn't appear to be a useful use of RAID 6 to me. You loose two
> disks to parity with RAID 6 (as I understand it) so with only four disks
> you would be better off striping and mirroring. Same capacity, 3TB and
> not performance lose in calculating two parity blocks.
I plan on growing the array as I start to fill it up.
Also as I understand it.
RAID 6 with 2 disk parity, I can lose any 2 disks in the array and still recover everything.
RAID 10 (which I think your talking about) would give better performance etc, and still give some
support for 2 disk failure. In the attached picture, if the 2 disks which failed where on the same
side of the tree i.e. 1 & 3 or 2 & 4 then you've lost everything, therefore it doesn't provide quite
as much resilience.
__
Sussex mailing list
Sussex at mailman.lug.org.uk
E-mail Address: sussex at mailman.lug.org.uk
Sussex LUG Website: http://www.sussex.lug.org.uk/
https://mailman.lug.org.uk/mailman/listinfo/sussex
-------------- next part --------------
A non-text attachment was scrubbed...
Name: raid10.gif
Type: image/gif
Size: 20966 bytes
Desc: not available
URL: <http://mailman.lug.org.uk/pipermail/sussex/attachments/20101109/3af55dd7/attachment-0001.gif>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 260 bytes
Desc: OpenPGP digital signature
URL: <http://mailman.lug.org.uk/pipermail/sussex/attachments/20101109/3af55dd7/attachment-0001.pgp>
More information about the Sussex
mailing list