Hi all, Jumping in on this thread late but (very bluntly) if your test harness throughput is dependent upon the response time of your application, you test system is broken. This is one of the conditions that we've recently named coordinated omission (CO). It is a state where the test harness and the server (accidentally or unintentionally) collaborate to miss injecting samples when they should have been injected. The net result is that the harness backs off allowing the server to drain and that in turn allows your server to report on numbers that are much better than they would be in reality. From a performance testing point of view, the reduced load is akin to a scale down study of your system. Scale down studies often produce different results than testing at scale. It is but one of the many ways UATs fail to recognize problems that leak out into prod but it's an important one.
How to fix this? For starters, always set the loop count for a thread to 1. Set the number of threads to the number of transactions you want to run and then set the warmup time to the duration of the test. This won't fix all of the problems that come with using JMeter but it gets rid of one of the bigger problems. BTW, all of the other tools in the space suffer from the exact same issues. Regards, Kirk On 2013-10-02, at 9:05 PM, Philippe Mouawad <[email protected]> wrote: > Hello, > Of course you can generate even much more than 21req / s. > If fully depends on your response times. > > Want to test: > - Create a mirror server listening on 8081 > - Create an HTTP Sampler hitting localhost:8081 > - Add only an aggregate report > > > Running on a last generation Mac Book Pro with default configuration => 517 > req/s . > > This test is stupid but just to confirm you can hit this rate. > The results you mention are strange and I think wrong. > > Regards > Philippe > @philmdot > > > > > > On Wed, Oct 2, 2013 at 12:47 PM, bobMeliev > <[email protected]<javascript:_e({}, 'cvml', '[email protected]');> >> wrote: > >> Nope. Even by default Dummy Sampler couldn't generate such load. Screenshot >> attached. >> >> <http://jmeter.512774.n5.nabble.com/file/n5718286/screenshot_233.png> >> >> 21 req/s is huge load just for 10 users. Check this report >> >> http://blazemeter.com/blog/increasing-productivity-wordpress-site-when-using-blazemeter-its-easy-task >> with 300 users generated only 5 req/s load and with such load CPU was 75% >> busy. >> >> >> >> >> -- >> View this message in context: >> http://jmeter.512774.n5.nabble.com/Understanding-Hits-per-second-tp5718263p5718286.html >> Sent from the JMeter - User mailing list archive at Nabble.com. >> >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: [email protected]<javascript:_e({}, >> 'cvml', '[email protected]');> >> For additional commands, e-mail: >> [email protected]<javascript:_e({}, 'cvml', >> '[email protected]');> >> >> > > > -- > Cordialement. > Philippe Mouawad. > > > > > -- > Cordialement. > Philippe Mouawad. --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
