[ILUG] Why RAID
paul at clubi.ie
Tue Jul 13 13:38:31 IST 2004
On Tue, 13 Jul 2004, John P. Looney wrote:
> Other benefits of RAID are speed. If you have two disks, in RAID1,
> you get twice the read performance, and a little over half the
> write performance (you can choose which disk to read from, but have
> to write to both). As 80% of the time you are reading, that's a
> plus. You can also configure them in RAID1+0, which means you get
> two or three pairs of disks, and mirror each pair, so that you get
> two or three "reliable" disks.
Ouch. RAID1+0 on four disks, so you get capacity and speed of
two-disk RAID0 but reliability of four disks where you can only
tolerate one failure? You'd be better off with RAID5 on three disks +
hot spare - sameish read performance, same capacity, but you can
tolerate two failures (provided second failure doesnt occur before
hot spare has synced).
Or RAID-5 on all four disks, probably better performance - same
Or try RAID-6, n-2 reliability. (but new)
> going can be time consuming. The next level up are the crappy pretend
> hardware RAID. Highpoint and Promise come to mind here. This can be a
> world of pain. Promise drivers are poor, and the machine still falls over
> when a disk dies
Yuk yes. Dont *ever* use the "RAID" features of these things, dont
*ever* let yourself be sucked into using their own drivers. On the
other hand, I believe it's possible for Promise (iirc) to setup
device-mapper mirror to grok their on-disk RAID format - bit of a
manual process at moment, hopefully in future there'll be tools to
automate it all. So you can use their BIOS RAID to boot, but use sane
linux drivers after that.
> You can then get little hotswap drive cages for SCSI (not seen it
> for SATA yet),
You can, I have a cage here. See www.acme-technologies.co.uk, Encom
> this can also be painful - we learned the hardway that if you put a
> certain brand of IBM disk into an Intel SR2300 without upgrading
> the disk's firmware, you can corrupt your RAID array.
Hardware RAID controllers sometimes do funny things, eg I've had
problems with Mylex controllers refusing to recognise a disk anymore
that had been pulled as good. The stubborn thing insisted the disk
had died, and by golly it wasnt going to recognise it anymore. I had
to change the SCSI ID to make it think it was a new disk - luckily
the drive was *not* in a cage (where IDs usually are assigned by the
cage). There was no other apparent way to make it work.
Paul Jakma paul at clubi.ie paul at jakma.org Key ID: 64A2FF6A
warning: do not ever send email to spam at dishone.st
The 357.73 Theory:
Auditors always reject expense accounts
with a bottom line divisible by 5.
More information about the ILUG