Summary: Oracle and NUMA and Tru64 v5.1

From: Robert Mulley <Robert.Mulley_at_unilever.com>
Date: Wed, 09 May 2001 16:08:07 +0100

Hello all,

Original question at end. Firstly thanks go to:
Bryan Williams
Michael Wheelock
Anokhin Stanislav
David Reed
David Hull
Greg Freemyer.

A dialogue with Bryan and some documents from Greg, David H and Michael
helped me sort out the problem in my head. The consensus was that I should
be running Oracle 8.1.7. Unfortunately it was out of my hands. It was like
that when I got here. In fact it was 7.3.4 at one stage. So as soon as
functional testing for 8.1.7 is done then we will move to that. Oracle
aside, the problem of slow commands was what had us really worried. A copy
of a 300M file could take from 15secs to 5.5mins.

As the problem seemed to be occuring with just about every unix command I
had to think of a way to solve it. These GS160's are very powerful and
configurable machines which can make them quite complex. Thus my
workaround won't work for everybody. What I ended up doing was setting up
a processor set that contained all the CPU's in the second RAD. This was
then allocated to our live Oracle database. We adjusted oracle's SGA to
fit into the 8G running there to cut down on cross Hswtich communication.
This seems to have sped everything back up to normal. Oracle isn't holding
up the unix processes which are running on the default processor set. This
isn't an ideal solution as Oracle is missing out on the use of 4 CPU's and
8G of memory.

I have been told that the way to get best performance is to hard partition
the machine, set up a cluster between the machines and run Oracle Parallel
Server. Due to management concerns this is not a possibility here.

Anyway most of this stuff won't make sense to most people on the list, one
person simply sent an email saying "Huh.". So if you have any questions,
comments etc... then please feel free to email me.

Robert.

Original question:
---------------------------------------
Peoples,

We are running Oracle 8.1.6.2 on a GS160 running v5.1 pk1. Our original
setup was one QBB (rad?) with 4 CPU's and 4G of memory. We upgraded to 2
full QBB's with 8 CPU's and 16G of RAM. We have plans to go to Oracle
8.1.7.1 but are unable to do this just yet.

Problem is that it looks like things are running slower than before.
Documents we've read suggest that the problem is due to the SGA (oracle db
buffers) being spread across multiple QBB's therefore relying to much on
the QBB H-switch. We've read lots of documents now and some of them
contradict each other. I would like to hear from people that are in the
same boat. Or have had this trouble and now fixed it. Also the wonderful
Compaq people that monitor this list, is there any definitive settings we
need to set to get this running properly.

TIA

Robert Mulley
Unix Admin

P.S. Problems with timing meant I couldn't install PK3 at the same time we
did the upgrade. This is planned to happen when we can get some more
downtime.
--------------------------------------
Received on Wed May 09 2001 - 15:10:12 NZST

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:42 NZDT