rh9 performance issue with oracle

[Thierry ITTY]

Joe,

thanks for the link

thought, i don't think the problem is only oracle related. I used oracle as
a bench tool (and also because it will be the main part of the application
these machines have to serve). Also, I (have to) use oracle 8.1.7
remember : the 2 machines have exactly the same oracle configuration, so
the difference is quite for sure os-related (filesystems, memory config). I
would hardly believe that a 2 way xeon with fwd scsi disks has lower
performance than a pIV with a single ide disk

my questions are for example : why is buffer storage so low ? what about
special tuning options for xfs filesystems ? which tools could I use to
compare latencies between the 2 machines and hopefully get some clues to
where the problem lies ? or which tests should i do (how to bench an IO
subsystem...)

TIA


A 11:11 02/12/2004 -0500, vous avez écrit :
>Have you seen this:
>http://www.puschitz.com/InstallingOracle9i.shtml
>
>There are links to tuning the kernel for faster Oracle.
>
>
>On Thu, 02 Dec 2004 17:01:41, Thierry ITTY <thierry.itty at besancon.org> wrote:
>> I have 2 machines :
>> - a mono-PIV with 512 MB, one 40 GB ide disk, RH9 2.4.20-8 and ext3
>> filesystems
>> - a bi-Xeon with 2.5 GB, two 14 GB fwd scsi disks, RH9
>> 2.4.20-9SGI_XFS_1.2.0smp and xfs filesystems
>> 
>> both have oracle 8.1.7.3 installed, with exactly the same database
>> (export/import)
>> 
>> we wrote a test script (timing a sql query executed 20 times) to compare
>> performance on the 2 machines
>> 
>> on the PIV machine the test runs in 0.8 seconds, and on the Xeon machine it
>> takes 1.1 seconds (many many tests, these are really definitive values)
>> 
>> I need some advice to investigate.
>> 
>> I read about "bounce buffering", "asynchronous io", shmmax values, and
so on,
>> One clue is that on the Xeon machine, cat /proc/meminfo always show very
>> little buffer usage (steady 60k) whereas on the other I've more than 100m.
>> 
>> I can't find anything helpful to solve this performance issue
>> 
>> any help greatly appreciated
>> 
>> TIA