Adrian Bridgett <adrian@???> wrote:
> a) RAID-5 + hot swap spare on four disks. So
> that's Data + Data + Parity + Spare - a 50%
> usage rate. RAID-1 will give you that and much
> better write performance. There is a higher
> chance of failure if you lose two disks though.
That's what I was thinking. This machine will be
installed in Eastleigh - best case, it will be
over an hour before I can get to it to replace a
failed disk. I want that extra insurance of a hot
spare, so I don't have to drop everything...
> b) I don't really think LVM will add that much
> overhead - yes it's a mapping but compared to
> disk performance.
No, I don't expect LVM would make much difference.
I am running all the site's services on a single
P3-800, though, so the RAID-5 calculations will be
quite enough extra processor load.
> In fact, the only thing I could see is about
> lvm1 (not lvm2) - I can't actually see any drop
> in performance!
> http://www.linuxkp.org/en/content.php?&content/server/lvm1.html#perf
This article doesn't say what spec system they're
using, although a 2001 Proliant 1850 probably
doesn't have a terribly fast processor.
They're taking a fairly simplistic view of the
SmartArray controller - I don't think it's
possible to totally disable them. If nothing else,
they usually have an i960 and substantial cache.
> http://arxiv.org/pdf/cs.PF/0508063.pdf
These tests were run on a P3-900 with 512MB RAM,
quite close to what I'm using. No RAID-5 tests,
unfortunately.
In the end, the decision not to use LVM was mostly
a case of "the kernel can partition RAID arrays,
so I'm going to use it." :-)
Nick.
--
Nick Chalk ................. once a Radio Designer
Confidence is failing to understand the problem.
