Re: Fault tolerance with RAID

From: Darren Dunham (ddunham_at_redwood.taos.com)
Date: 08/02/03 

Date: Sat, 02 Aug 2003 17:40:57 GMT

Phil Meyer <x0pmeyer@ti.com> wrote:
> Greg wrote:

> In terms of making the OS itself as safe as possible from disk outages,
> please reconsider the use of any software based raid. Whether Veritas
> encapsulated root disks or SDS mirrored root disks you run as great a risk
> of data corruption from the mirroring software as you do a catastrophic
> disk failure.

If the disks are going to give you silent failures, why would a
"hardware mirror" be any less susceptible?

> I have personally witnessed this very scenerio with both Veritas and
> Disksuite. It is a very sinking feeling you get as you watch a critical
> system eat itself to the point of no return.

I don't see anything that makes an external array immune to such issues.

I personally think that..

1) silent corruption like this is relatively rare.
2) "non-silent" disk failures are less rare.

Given that, I would rather have *some* protection (sofware mirrors)
rather than nothing else.

> An additional oddity with SDS is that it is very possible to replace a
> failed disk, partition it correctly, and enable it, to watch SDS sync the
> wrong way! Thus corrupting the good disk with whatever was on the new
> disk. Not fun.

Ouch. I've never seen that. Is there an open bug on that?

> It really depends upon your Service Level expectations, but software based
> raid is only about 40% effective (In My Experience), and the other 60% of
> the time it IS the problem. :)

Your numbers are very different from mine. 

-- 
Darren Dunham                                           ddunham@taos.com
Unix System Administrator                    Taos - The SysAdmin Company
Got some Dr Pepper?                           San Francisco, CA bay area
         < This line left intentionally blank to confuse you. >

Relevant Pages