On 19/09/2007, Christiaan Lamprecht <[EMAIL PROTECTED]> wrote: > Hey all,
Which version of JMeter? > I'm trying to find out exactly what Elapsed Time includes, it seems to > have a big impact on my experiments as I'm getting some funny results > and one of the machines keep on crashing. Latency results on the other > hand seems fine. Any help would be very much appreciated. Elapsed time is total time for the sample, i.e. from initiating the request to its completion. > > Experiment: > > Test Plan > Thread Group (users 5, ramp 0, count 250) > HTTP Request HTTPClient (SSL request) > Simple Data Writer > Thread Group (users 10, ramp 0, count 250) > HTTP Request HTTPClient (SSL request) > Simple Data Writer > Thread Group (users 15, ramp 0, count 250) > HTTP Request HTTPClient (SSL request) > Simple Data Writer > Thread Group (users 20, ramp 0, count 250) > HTTP Request HTTPClient (SSL request) > Simple Data Writer > Thread Group (users 25, ramp 0, count 250) > HTTP Request HTTPClient (SSL request) > Simple Data Writer > Thread Group (users 30, ramp 0, count 250) > HTTP Request HTTPClient (SSL request) > Simple Data Writer > Thread Group (users 35, ramp 0, count 250) > HTTP Request HTTPClient (SSL request) > Simple Data Writer > Thread Group (users 40, ramp 0, count 250) > HTTP Request HTTPClient (SSL request) > Simple Data Writer > Gaussian Random Timer (Delay offset 100, deviation 20) Are these threads run in parallel, or in sequence? Why is ramp-up zero? Are you using CSV or XML output? > > 6 identical machines; 1 server, 4 clients and 1 collecting the > results. All on the same local network making secure requests for a > single 131K pdf file. > > > Issue 1: > Upon inspecting the results files (the request timestamps) it seems > that my request rate is actually decreasing over time from about 55 > req/s to 12 req/s! and then the machine collecting the results crashes > but the others, including the server seems fine. The client JMeter process has to handle all the response traffic. Are you using any of the deferred response options to try and reduce the load? Or you could try using independent non-GUI tests and combine the results manually at the end. > Issue 2: > > The Elapsed time INcreases as follows: > > ----------- Elapsed time ----------------- > Mean: 145.9614 > Confidence interval: 4.4535686733307775 > > ----------- Elapsed time ----------------- > Mean: 271.8111 > Confidence interval: 10.990474145938794 > > ----------- Elapsed time ----------------- > Mean: 441.96093333333334 > Confidence interval: 15.121548663364933 > > ----------- Elapsed time ----------------- > Mean: 577.16795 > Confidence interval: 17.569224709437265 > > ----------- Elapsed time ----------------- > Mean: 784.111 > Confidence interval: 23.90328057326669 > > ----------- Elapsed time ----------------- > Mean: 738.2100486223663 > Confidence interval: 30.158344351290822 > > ----------- Elapsed time ----------------- > Mean: 1723.2670157068062 > Confidence interval: 348.28728611889505 > > > > > > But the latency DEcreases towards the end: > ----------- Latency ----------------- > Mean: 57.7488 > Confidence interval: 3.7233761673353256 > > ----------- Latency ----------------- > Mean: 117.8008 > Confidence interval: 7.795127906431532 > > ----------- Latency ----------------- > Mean: 187.74713333333332 > Confidence interval: 7.9288495213299 > > ----------- Latency ----------------- > Mean: 228.9634 > Confidence interval: 7.752167562857181 > > ----------- Latency ----------------- > Mean: 309.96296 > Confidence interval: 9.900207087746896 > > ----------- Latency ----------------- > Mean: 285.04105888708807 > Confidence interval: 14.377120160947742 > > ----------- Latency ----------------- > Mean: 101.07329842931937 > Confidence interval: 37.688563788937685 > > > > What usually happens is that over time the server response time How do you measure this "usually" ? > increases steadily and then caps off (i.e. does not increase further) > at a certain client load regardless of further request increases. The > latency numbers might be showing this if it was allowed to run for a > bit longer, if only the Elapsed Time didn't peak. > > > Any help would be very much appreciated! Overloading of the client JMeter process may be causing some or all of the results you are seeing. > > Thanks again > Christiaan > > --------------------------------------------------------------------- > 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]
