On 09/09/2009, Marc Chiarini <[email protected]> wrote: > Hi Folks, > > I am trying to verify that I am looking at my "problem" from the correct > perspective. I have a single thread group and a http sampler that > requests the same file repeatedly for some user-defined duration (say > 300 seconds). The sampler has a Gaussian Random Timer as a child with > offset and std dev set manually (e.g., 500ms offset and 50ms deviation), > in order to rule out any weirdness using variables. When I run the test > with one thread, I obtain my desired behavior: the delay between > requests is normally distributed with mean and std dev very close to the > timer settings. However when I use 2 or more threads, I get delay > distributions that are not fitting any particular distribution and are > certainly not normal. As I go higher with the number of threads (and > correspondingly increase the offset and std dev, just to make sure the > center mass is sufficiently distant from zero), the distribution starts > to look exponential. Am I missing something? If it is true that the > sum of two or more normal distributions is itself a normal distribution, > then I don't understand why using multiple threads in this fashion > results in this behavior. > > Any help is greatly appreciated.
Try using a Java Request sampler instead of the HttpSampler. That will eliminate any network or server variations, and will prove a useful test case in case there is a bug in JMeter. Obviously you won't be able to test the server, but it should make investigating the timing issues much simpler. > Regards, > Marc > > PS I reproduced the generation of normally distributed delay times > using a Beanshell timer, but obtain similar results. > > > > --------------------------------------------------------------------- > 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]
