Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Compaq NUMA

Compaq NUMA

From: Andrew Mobbs <andrewm_at_chiark.greenend.org.uk>
Date: 23 Nov 2001 13:04:18 +0000 (GMT)
Message-ID: <2fy*uj5-o@news.chiark.greenend.org.uk>


I seem to ask this here every few months when one of these machines comes back to haunt me.

Has anybody managed to get good performance, and good scaling running a single instance of Oracle on a Wildfire class machine? How? Any tips at all are appreciated. We've tried using runon to bind processes to RADs, which gives something, but still not really enough.
(Ob. version numbers - 8.1.7.2, Tru64 5.1A)

Compaq's TPC-C tests all run with OPS (or RAC) clustering separate instances running on each node. Which nicely gets around the NUMA problem for a benchmark, is somewhat impractical and expensive for a productions system.

One other thing - in their latest GS320 TPC-C benchmark, Compaq seem to have set
_wait_for_sync = FALSE
(See Full Disclosure report page 219 -

http://www.tpc.org/results/FDR/TPCC/compaq.gs320.v5.062501.fdr.pdf )

SQL> select KSPPDESC from X$KSPPI where KSPPINM = '_wait_for_sync';

KSPPDESC



wait for sync on commit MUST BE ALWAYS TRUE

Well, that's consistent, durable transactions thrown out in the name of performance, unless there's something special about LGWR on Tru64? This doesn't seem to be usual, even in the world of competitive benchmarking.

-- 
Andrew Mobbs - http://www.chiark.greenend.org.uk/~andrewm/
Received on Fri Nov 23 2001 - 07:04:18 CST

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US