Thanks for the quick reply Jordi.

I'll have to go back to make sure, but I was monitoring the jmeter
server(s) at the same time and I don't remember max'ing them out. Damn,
I should have captured those stats too (I have 20 pages of stats from
the web and database server).

BTW, how can you tell it was max'ed at 1x20? Hm. Then I guess anything
larger than 20 threads in my tests is crap? Oddly enough, the load on
the database side kept increasing with the number of threads. I went up
to 80 threads on a previous test and almost killed the database server.
:-)

Thanks!
js.

On Wed, Oct 01, 2003 at 05:20:34PM +0200, Jordi Salvat i Alabart wrote:
> Looks like around 1x20 you saturated the JMeter CPU. 1x40 did not 
> increase load -- not because the web server was saturated, but because 
> the JMeter machine was. When you did the 2x20, you were actually 
> providing twice as much CPU power to JMeter, so you could create twice 
> as much load.
> 
> Jordi.
> 
> Jean-Sebastien Morisset wrote:
>> Guys,
>> 
>> I ran some tests overnight with jmeter v1.9.1, and got some odd results...
>> 
>> --------------------------------------------------------------------------
>> 
>> Threads      Rate    R/Thr.  Avg     CPU
>> 1 x 1        13/sec  13/sec  75ms    28%
>> 1 x 5        22/sec  4.4/sec 222ms   38%
>> 1 x 10       23/sec  2.3/sec 436ms   37%
>> 1 x 15       23/sec  1.5/sec 646ms   39%
>> 1 x 20       23/sec  1.15/sec    850ms   38%
>> 1 x 30       24/sec  0.8/sec 1251ms  38%
>> 1 x 40       24/sec  0.6/sec 1665ms  38%
>> 2 x 20       57/sec  0.7/sec 1265ms  72%
>> 2 x 30       57/sec  1.4/sec 1541ms  72%
>> 2 x 40       57/sec  0.7/sec 1486ms  72%
>> 2 x 60       57/sec  0.5/sec 2220ms  72%
>> 
>> * CPU is the web server CPU.
>> 
>> * The Threads # x # is the number of jmeter servers x the number of
>> threads on each.
>> 
>> * When the jmeter servers were increased to 2, I added the Rates of each
>> together to get a total # of requests/second.
>> 
>> --------------------------------------------------------------------------
>> 
>> Notice one of the tests was 1x40 and another 2x20. These two tests should
>> have given the same results since the total number of threads is the same.
>> 
>> When using 2 jmeter servers, I added the Rate together to get a total
>> number of requests / second. This _seems_ logical since each jmeter
>> server is telling me it can do X number of requests per second, but the
>> 1x40 and 2x20 tests should then have the same Rate number.
>> 
>> I also don't understand why the web server's CPU jumped from 38% -> 72%
>> when going from 1x40 to 2x20. This would indicate twice the number of
>> users/threads.
>> 
>> Any thoughts? :-)
>> 
>> Thanks,
>> js.
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [EMAIL PROTECTED]
> For additional commands, e-mail: [EMAIL PROTECTED]
> 

-- 
Jean-Sebastien Morisset, Sr. UNIX Administrator <[EMAIL PROTECTED]>
Personal Home Page <http://jsmoriss.mvlan.net/>
JS & Melanie's Homebrewery <http://brewery.mvlan.net/>
Underwater and Travel Photographs <http://www.mvpix.com/>

---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to