Re: Best utility to max out CPU usage by DB workload in test environment?

I will need to look into this re: Calibrate IO - I looked into it quite deep at the time (dtrace, various oracle and RH tools, fiber channel and array tools) and found it was only using one instance . It may have been on 19.10.

On Tue, Nov 19, 2024 at 7:47 AM Mikhail Velikikh <mvelikikh_at_gmail.com> wrote:

> Calibrate_IO is another good and simple IO stress test, but it is *limited
>> to a single host in a clustered / Exa environment* , whereas SLOB is
>> better for Exa / RAC workloads.
>
>
> Calibrate_IO uses all instances in a clustered setup. It should be the
> same on Exa.
> This is documented:
> https://docs.oracle.com/en/database/oracle/oracle-database/19/arpls/DBMS_RESOURCE_MANAGER.html#GUID-90D0674E-6A10-41A9-ABBD-783F908502C3
>
>> For an Oracle Real Application Clusters (Oracle RAC) database, the
>> workload is simultaneously generated from all instances.
>
>
> There was a bug 30013535 when Calibrate_IO used a single instance only (DBMS_RESOURCE_MANAGER.CALIBRATE_IO
> Not Work Well On RAC 18c / 19c (Doc ID 2703766.1)
> <https://support.oracle.com/rs?type=doc&id=2703766.1>).
> The fix for the bug is included in 19.11.
>
> -
>
>
> On Mon, 18 Nov 2024 at 20:44, Jon Crisler <joncrisler_at_gmail.com> wrote:
>
>> I am not sure SLOB will achieve what Kunwar seeks. Pushing heavy IO
>> will stress the storage paths, buss adapters /HBA or network cards, but may
>> not affect CPU as much as hoped. I was able to successfully model heavy
>> IO combined with high cpu by using both SLOB and a cpu stress utility in
>> tandem. Stress or Stress-ng works well, but there are lots of cpu stress
>> tests available. Another question we had to answer was "will our
>> network/firewalls keep up with application traffic" between app jvms and
>> Oracle db. This was done by customizing Swingbench to put heavy emphasis
>> on network traffic. Both of these tests were successful in predicting that
>> we would encounter both storage I/O problems and firewall problems.
>> Storage was fixed with array tuning, adding a few more io paths. Firewall
>> was more difficult and we had to change the platform, then do extensive
>> tuning.
>> Calibrate_IO is another good and simple IO stress test, but it is
>> limited to a single host in a clustered / Exa environment , whereas SLOB is
>> better for Exa / RAC workloads.
>>
>> On Mon, Nov 18, 2024 at 6:15 AM Stefan Koehler <contact_at_soocs.de> wrote:
>>
>>> Hello Kunwar,
>>> I would go with SLOB (
>>> https://github.com/therealkevinc/SLOB_distribution/tree/2.5.3 ) and run
>>> it for LIO load only.
>>>
>>> Best Regards
>>> Stefan Koehler
>>>
>>> Independent Oracle performance consultant and researcher
>>> Website: www.soocs.de
>>> Twitter: _at_OracleSK
>>>
>>> > kunwar singh <krishsingh.111_at_gmail.com> hat am 17.11.2024 23:09 CET
>>> geschrieben:
>>> >
>>> > Hey Oracle experts,
>>> > What utility and flags do you suggest to quickly load Database CPU
>>> utilization to 100%?
>>> > Say I have cpu_count=16 on a 64 core machine and I want the database
>>> to use up all 16 cpus.
>>> >
>>> > For example I use stress-ng to get high CPU utilization on a linux box.
>>> >
>>> > I am playing with swingbench but so far no luck.. Or maybe I am using
>>> it the wrong way :)
>>> >
>>> > --
>>> >
>>> > Cheers,
>>> > Kunwar
>>> --
>>> http://www.freelists.org/webpage/oracle-l
>>>
>>>
>>>

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