From: NetComrade (netcomrade@bookexchange.net)
Date: Thu Jun 01 2006 - 02:51:28 EDT
My apologies as this is a bit OT, but I have nowehere else to turn, and
EMC has so far been completely unhelpful (btw, anyone know if they have a
concept of 'account rep' and 'technical rep', as I just joined a new
company and they have a dozen of various clariions, and I figure I should
be able to leverage that (threaten with HDS or Pillar :) with support). I
have also demanded escalation for 2 biz days now, and so far I've been
told: "I.m going to try to open a DIMs with engineer to see what is going
on with the performance group "
I've had some experience in the past migrating and growing Veritas file
systems. The experience I had was on old Sparc II hardware running
VM3.1.. I also thought of Sparc II as being a bottleneck, and thought the
biuld rates were slightly slow.. When the same disk arrays were attached
to Sparc III, sequential IO was about 30% faster, which seemed to confirm
my 'sparc II is the bottleneck' suspicions.
So now we have an EMC CX300 which, as far as I know, runs fairly fast Xeon
processors, and should be way faster then the VM on Sparc II.. Generally
it seemed that way, as we have a Veritas cluster running Oracle on EMC
CX500 and IO seems noticeably faster (although a SATABoy that we have
seems to outperform it in sequential IO, at least in datafile creation
times).
Anyway, I was trying to move some LUNs around on a CX300, and it seems
painfully slow. A 20G to a 60G LUN migration within a 6 disk RADI10 with
no IO going in ASAP mode took about an hour. The same migration with
decent IO going on (had a script to copy /usr onto the LUN in a loop) took
way over 7hrs on fibre drives with no other activity going on on the
array.
EMC is giving me some bul: it might be slow for this and that reason, and
they don't have an table with 'estimates' how much certain sizes will take
to migrate.
I know this is not an EMC list, but I wanted to get a feeling from those
that have done this enough on Veritas and/or EMC or other products, and
maybe get some feedback from those that worked on both and have had
similar frustrations. (e.g., how long would a similar migration take with
Veritas on a host with fairly fast CPUs? Is EMC not really capable of
doing online migrations vs. Veritas?)
The migration seems to go on fine (with no performance hit on CX300) until
the 'old LUN'/'new LUN' threshold is hit.. E.g. when migrating 30G to 60G
LUN, I get a huge (10x) performance hit at 50%, when migrating 20G to 60G
LUN I get a similar performance hit at ~33% completion. I have tried a
'lighter' LUN load (e.g. copying a 20meg directory in /usr every 20
seconds), but that didn't seem to 'improve' things dramatically.
Could this be CX300 specific? We really are delaying a LUN migration on a
CX500 b/c of this.
Thanks for any feedback, and once again, my apologies if it seems totally
OT.
-Andrey
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 23:40:00 EDT