>> On Fri, 20 Jan 2006 14:02:28 -0500, David Longo <David.Longo AT HEALTH-FIRST 
>> DOT ORG> said:


> I've just done a reconfig of my DB and LOG volumes that flies in the
> face of conventional wisdom - but it works!

[...]

I think that most TSM admins tend to prefer as many ways to keep their
pants up as can be arranged.  I've mused in the past that, if IBM
allowed 4 copies of a volume, than three would very quickly become de
rigeur for standard running.  (you -need- the unallocated copy if
you're going to migrate to new disk tech)

For a long time I used as many spindles as I could, to free database
work of contention.   With disk tech such as you are using now, and
huge caches reliably inserted between you and the disk, the strength
of that need has dropped precipitously.

RAID on the back-end ameliorates many (but as has been pointed out,
clearly not all) of the risks which the TSM-level mirror was intended
to address.  As a result, your 2:1 cost penalty for mirror disk
allocation (mumble RAID overhead, mumble hot spare overhead) is being
deployed to chase a smaller and smaller risk.  It is entirely rational
to conclude that, in your environment it is "too much" cost for "not
enough" coverage.

But I would very strongly suggest that you put numbers by that cost,
and try to estimate the risk, too.  Ever do dimensional analysis in
physics or chemistry?

Re-express 5 SATA disks in hours-of-service-down. :) I'm keeping the
mirrors.


- Allen S. Rout