Re: [RAC 11gR2] stress tool
From: TheBoss <TheBoss_at_invalid.nl>
Date: 07 Aug 2013 22:33:16 GMT
Message-ID: <XnsA2165A4B92E3TheBossUsenet_at_213.75.12.10>
joel garry <joel-garry_at_home.com> wrote in news:49c9c813-0058-4dc5-be62-56f2fb4cad14_at_googlegroups.com:
>> Mladen Gogala <gogala.mladen _at_gmail.com> writes:
>>
>>
>>
>> > On Mon, 05 Aug 2013 00:06:38 +0200, noreply wrote:
>>
>> >
>>
>> >> Hello,
>>
>> >>
>>
>> >> I was charged to stress a 3-nodes 11gR2 RAC database last week. I
>> >> have
>>
>> >> tried many approaches in order to detect the "breakpoint" but I
>> >> failed.
>>
>> >>
>>
>> >> I never reached this limit and thus, I could not conclude anything
>> >> about
>>
>> >> the current setup.
>>
>> >>
>>
>> >> This is the first time I try to "benchmark" a 11gR2 RAC database
>> >> and I
>>
>> >> must confess, I have no idea how to break it down.
>>
>> >>
>>
>> >> What tool could I use to reach the breakpoint ? What method would
>> >> you
>>
>> >> use to do so ?
>>
>> >>
>>
>> >> The nodes are all like this: HP-UX 11iv3 - 8 cores - 16G RAM. 10Gb
>>
>> >> interlink (UDP).
>>
>> >>
>>
>> >> Thank you very much
>>
>> >
>>
>> > Now, that's a simple task: create two tables, in a parent->child
>>
>> > relationship (foreign key) and update them from all 3 nodes
>>
>> > simultaneously, doing a mix of inserts, updates and deletes. Make
>> > sure
>>
>> > that all 3 nodes do work on the same blocks. It will break it, no
>> > matter
>>
>> > how powerfull the nodes are.
>>
>>
>>
>> Ah exact. This is not really what I need in fact ^^. Maybe my english
>> is
>>
>> not good to describe it correctly.
>>
>>
>>
>> Let my try again.
>>
>>
>>
>> What I would like to test is the throughput of the system. I need
>>
>> to show how powerfull the system is and how many TPS it can support
>>
>> before it will break.
>>
>>
>>
>> I agree this is all dependent on the SQL code being written by the
>>
>> developers and this can't be taken as a "proof".
>>
>>
>>
>> I just need to stress it until it break in a "theorical sense" -i.e
>>
>> placing the environment as if it was the best conditions it could
>> ever
>>
>> face - (a little bit like in an Oracle Lab ;)) and report that number
>>
>> (#user, #tps, ...).
>>
>>
>>
>> Hope it is clearer :)
>>
>>
>>
>> Thank you
>>
>> --
>>
>> XMA
Date: 07 Aug 2013 22:33:16 GMT
Message-ID: <XnsA2165A4B92E3TheBossUsenet_at_213.75.12.10>
joel garry <joel-garry_at_home.com> wrote in news:49c9c813-0058-4dc5-be62-56f2fb4cad14_at_googlegroups.com:
> On Monday, August 5, 2013 1:40:15 PM UTC-7, nor..._at_maillard.im wrote:
>> Mladen Gogala <gogala.mladen _at_gmail.com> writes:
>>
>>
>>
>> > On Mon, 05 Aug 2013 00:06:38 +0200, noreply wrote:
>>
>> >
>>
>> >> Hello,
>>
>> >>
>>
>> >> I was charged to stress a 3-nodes 11gR2 RAC database last week. I
>> >> have
>>
>> >> tried many approaches in order to detect the "breakpoint" but I
>> >> failed.
>>
>> >>
>>
>> >> I never reached this limit and thus, I could not conclude anything
>> >> about
>>
>> >> the current setup.
>>
>> >>
>>
>> >> This is the first time I try to "benchmark" a 11gR2 RAC database
>> >> and I
>>
>> >> must confess, I have no idea how to break it down.
>>
>> >>
>>
>> >> What tool could I use to reach the breakpoint ? What method would
>> >> you
>>
>> >> use to do so ?
>>
>> >>
>>
>> >> The nodes are all like this: HP-UX 11iv3 - 8 cores - 16G RAM. 10Gb
>>
>> >> interlink (UDP).
>>
>> >>
>>
>> >> Thank you very much
>>
>> >
>>
>> > Now, that's a simple task: create two tables, in a parent->child
>>
>> > relationship (foreign key) and update them from all 3 nodes
>>
>> > simultaneously, doing a mix of inserts, updates and deletes. Make
>> > sure
>>
>> > that all 3 nodes do work on the same blocks. It will break it, no
>> > matter
>>
>> > how powerfull the nodes are.
>>
>>
>>
>> Ah exact. This is not really what I need in fact ^^. Maybe my english
>> is
>>
>> not good to describe it correctly.
>>
>>
>>
>> Let my try again.
>>
>>
>>
>> What I would like to test is the throughput of the system. I need
>>
>> to show how powerfull the system is and how many TPS it can support
>>
>> before it will break.
>>
>>
>>
>> I agree this is all dependent on the SQL code being written by the
>>
>> developers and this can't be taken as a "proof".
>>
>>
>>
>> I just need to stress it until it break in a "theorical sense" -i.e
>>
>> placing the environment as if it was the best conditions it could
>> ever
>>
>> face - (a little bit like in an Oracle Lab ;)) and report that number
>>
>> (#user, #tps, ...).
>>
>>
>>
>> Hope it is clearer :)
>>
>>
>>
>> Thank you
>>
>> --
>>
>> XMA
> > You might want to google Oracle SLOB. >
You're outdated, it's SLOB2 now! ;-)
Have a look at this very educational blog post: http://flashdba.com/database/useful-scripts/using-slob2-for-pio-testing/
Jeff Moss posted additional info (and adapted script) specific for RAC: http://www.oramoss.com/blog/rac-aware-slob2-analyze-script-flash-dba/
(script: http://www.oramoss.com/wiki/index.php?title=slob2_analyze_rac )
Don't know why though, as you probably don't need RAC, right? ;-) http://flashdba.com/2012/09/10/database-virtualisation-the-end-of- oracle-rac/
For some alternative load testing tools: http://www.pythian.com/blog/oracle-database-load-testing-tools-overview/
Cheers!
-- JeroenReceived on Thu Aug 08 2013 - 00:33:16 CEST