what i understood from experts discussion is that the closer the idea of
its disk subsystem the os has matches the physical architecture, the better
the global performance you get. 

it's obvious with a flat, non raid system.

with a storage array, i'd say there's no use to create small luns if they
just divide one raid unit (i don't know about shark's architecture, just
working with hitachi). thought, it could make sense if the raid unit you
divide can handle several requests at the same time (that is, it has
several internal pathes to the physical disks)

for example, i've been advised (by hitachi) that, if i need one big logical
volume, i would get quite a better performance with only one big LUN
managed by the array natively instead of grouping several smaller LUNs in
on logical volume at the os level.

hth


A 21:01 05/11/01 +1300, vous avez écrit :
>Is there any advantage in creating multiple LUNs from a SAN to an AIX
>host for stgpool volumes?
>
>I have created a single large LUN on our SAN and put a bunch of backuppool
>volumes within the filesystem.
>
>As the system is growing I'm finding the need to do some tuning (as things
>are not as springy as they once were). I'm seeing a lot more Wait I/Os and
>things like migrations have definately slowed down.
>
>It got pointed out to me by a friend who said it was better to create
>a bunch of smaller LUNs, as it spread the load on the Storage array end
>(fair enough). But he also said it slowed the system down due to volume level
>write locking.
>
>I didn't think modern Unix systems used volume level write locking
>(understandable of the physical disk write constriction, but not from the
>OS level)
>
>The storage array in question is a Shark, I created the LUN from a complete
>rank (200GB). The peak disk throughput from the TSM system is around
30MB/sec.
>When the tape drives are firing to each other, it can almost be double on the
>HBA.
>The Shark gets hit more during the day from a bunch of DB servers, plus it's
>only 1.6TB populated. The last time we had the ESS expert running, there was
>95% read and 100% write to cache. The load hasn't significantly increased.
>
>Any filesystem experts willing to share their opinion?
>
>Cheers, Suad
>--
>
>
                        - * - * - * - * - * - * -
Si nous avons chacun un objet et que nous les echangeons, 
   nous avons encore chacun un objet.
Si nous avons chacun une idee et que nous les echangeons,
   nous avons alors chacun deux idees.

Thierry ITTY
eMail : Thierry.Itty AT Besancon DOT org               FRANCE