Re: Performance problem

From: Green, Simon (Simon.Green_at_EU.ALTRIA.COM)
Date: 08/03/05

  • Next message: Vincent D'Antonio: "Re: Off Topic -- TSM" 
    Date:         Wed, 3 Aug 2005 15:36:37 +0200
To: aix-l@Princeton.EDU

    You're doing an awful lot of paging. I think that's where you should target
    your efforts. Your system is spending so much time paging in and out that
    there's no time to actually do any work!

    Do you have statistics from before the performance problem, for comparison?

    Is there any new work going on? A user suddenly running a new query,
    perhaps? Or has a developer switched on a debugging option?

    What's your paging space configuration? It won't address the underlying
    problem but may alleviate it if you can improve paging performance.

    Your vmtune settings look OK, but you could reduce minperm and maxperm. We
    typically use -p 3 -P8 for large servers. We also specify more bufstructs.
    (-b 200, or -b 400.)

    We also tend to use a lot more aio servers, but I don't think that is your
    problem at the moment. 

    -- 
Simon Green							
Altria ITSC Europe s.a.r.l.
AIX-L Archive at https://new-lists.princeton.edu/listserv/aix-l.html
New to AIX? http://publib-b.boulder.ibm.com/redbooks.nsf/portals/UNIX
N.B. Unsolicited email from vendors will not be appreciated.
Please post all follow-ups to the list.
-----Original Message-----
From: IBM AIX Discussion List [mailto:aix-l@Princeton.EDU] On Behalf Of
Fernandez Garay Jorge
Sent: 02 August 2005 21:33
To: aix-l@Princeton.EDU
Subject: Performance problem
I wonder if someone in the list can help us determining the causes of this
problem.
Perhaps someone can, at least, direct us to look in some direction.
Any help will be appreciated
Thanks in advance
Jorge 
We have an S80 box (AIX 4.3.3), loaded with intensive database applications
(Oracle 8i).
12 Processors
10 GB RAM
Data Base Datafiles located at a Massive Storage device (EVA 5000 from HP).
None raw devices,
(Normal JFS filesystems)
The Databases are configured to perform Asynchronous I/O
Some time ago, we have tunned VMM with following command
/usr/samples/kernel/vmtune -p5 -P20 -f720 -F768 -W256
It had been working fine, till past friday. 
sar command output shows a heavy %wio (50-60%)
vmstat command output shows large values for wait queues (65-70)
We couldn't identify any additional workloads.
We have added 6 processors, and enlarged the 'maxserver' parameter for
Asynchronous I/O, from 120
to 180 ( 10 x Number of Processors)
New 'minserver' parameter value was enlarges from 60 to 90 (New 'maxserver'
value/2)
After that, things are getting far from better.
Here is a sample of sar command output
sar 1 50
<SNIP>
  • Next message: Vincent D'Antonio: "Re: Off Topic -- TSM"