RE: RAC in NAS

From: Kevin Closson <kevinc_at_polyserve.com>
Date: Thu, 27 Jul 2006 08:10:09 -0700
Message-ID: <5D2570CAFC98974F9B6A759D1C74BAD0E5ABEB@ex2.ms.polyserve.com>

>>>

>>>Asynch I/O is not that important. Oracle can emulate it
>>>using I/O slaves. Granted, it's not as good as the real
>>>thing,

...right, Mladen, it isn't. I wouldn't recommend slaves as musch as just multiple true writers. With 10g would can have a more than 10 of them. I routinely test 4 node Linux RAC on NAS with OLTP workloads that generate ~15,000 I/O sec where ~1000 are LGWR writes and 6,000 are dbwr writes. In the end if free buffer waits aren't the problem, all you are doing is improving (slightly) processor utilization when you use OS async IO.

Large SMPs generally benefit more from async IO than commodity servers, that is true (and I know why, and would love to have the time to explain why), but most large SMPs out there are servicing workloads that aren't as I/O intensive as the ones I test PolyServe Scalable NAS with (HP EFS-CG). Just look at the numbers.

--
http://www.freelists.org/webpage/oracle-l

Received on Thu Jul 27 2006 - 10:10:09 CDT