[plug] Raid 0 software? hardware?

[Craig Ringer]

On Sat, 2005-04-02 at 20:32 +0800, Cameron Patrick wrote:

> I believe there's a "dmraid" driver which provides a layer on top of
> the standard Linux software RAID system, allowing it to access data
> stored in the proprietary format that the "RAID" card firmware uses.
> I've never used it so I'm not sure how it compares to pure Linux
> software RAID, but I'd imagine it would be as least as good as any
> drivers supplied by the card vendor.

If it's anything like the older 'md' driver, I imagine it'd be pretty
good.

> > I'm also kind of concerned that you mention that you're using RAID 0. 
> > RAID 0 across four disks is a death sentence to your data, even with 
> > high quality "enterprise" SCSI/SATA disks. With "consumer" level 
> > ATA/SATA disks I'd expect total loss of all data on the array within a 
> > year - at least going by my luck with disks.
> 
> It's worth noting that you should keep backups even if you are using
> one of the "reliable" RAID levels. 

I couldn't agree more. Double disk failures happen. People's servers get
stolen or catch fire. OSes go insane and decide that writing gibberish
over the disk is fun. Filesystems get corrupted. RAID controllers fail
and start corrupting writes. RAID won't save you from any of these.

> However losing 1 in 4 drives in a
> year sounds a lot worse than my luck with cheapo ATA drives.

I agree - I still don't understand it but it just keeps on happening.
I've had particularly terrible results with Western Digital disks. I've
had three 120GB WDs at home - two of which have died. One was a warranty
replacement for a previously failed WD 120GB. Each failure happened in a
different system and both were well cooled.

I've had so many WD disks die at work that I just call Austin and say
"got another dead one, I'll send it in with the courier for the next
order". Of the original 3 disk 120GB WD RAID array I had three fail (I
was later able to confirm I got a defective batch of disks). The server
now has an array of five 250GB WDs in it (4xRAID 5 + hot spare). I think
I've lost three of those over time now. The server is well cooled and
the disks are not under exceptionally heavy loads, so unless vibration
is killing them (and there's not that much vibration...) or something I
just don't know what could be going on. The 2x80GB Seagate disks the OS
RAID 1 sits on have been dead reliable the entire time, as has the
Maxtor in my home machine.

So ... with a track record like this, yes, I do assume disks will die
and die quickly. This track record on disk reliability isn't confined to
any particular system or site and doesn't appear to be to do with
cooling or power problems. The only pattern is disk manufacturer.

Even with "normal" disk reliability, I wouldn't be too surprised to see
one disk out of four die within a year. It's certainly not the sort of
chance I'd risk anything I cared about even slightly on. I'd be
reluctant to use RAID 0 even across "enterprise" SCSI disks, frankly ...
it's just asking for trouble. RAID 0 is great for things like giant
working areas for video editing, but I wouldn't use it to store anything
important, ever.

-- 
Craig Ringer