On 26/02/2010, [email protected] <[email protected]> wrote: > In gmane.comp.jakarta.jmeter.user, you wrote: > > In gmane.comp.jakarta.jmeter.user, you wrote: > >> On 26/02/2010, Doki Pen <[email protected]> wrote: > >>> I'm seeing horrible jmeter performance when I run in master slave mode > >>> on EC2 (10pages/sec). But when I run in normal mode I get great results > >>> (200 pages/sec) > >>> > >> > >> What about the individual response times? > >> Are they similar? > >> > >> Are the tests retrieving identical responses? > >> > > > > In the normal way of the internets, I discovered the solution to my > > problem 5 minutes after asking a usergroup. This, after 2 days of > > struggling. It turns out the magically incantation is as follows: > > > > # jmeter.properties > > mode=Batch > > hold_samples=true > > num_sample_threshold=1000 > > time_threshold=30000 > > > > > After more expirimenting, this isn't really what I want. hold_samples > makes each slave hold all samples in memory until the test is over, > usually causing on OutOfMemory error. num_sample_threshold and > time_threshold are what I needed. But it's a black art setting them, > because if they are too small, you get bad perf. If they are too big, > you run out of memory. >
You could perhaps try the statistical sampling mode. If that does not have enough detail, then just use independent non-GUI runs and combine the data after the runs complete. > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > > --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
