The View Result in Table Listener is fairly expensive as it has to keep details for every sample. Try disabling it. If you use CSV output you can reload the data later.
However, I think the problem is purely due to the excessively short ramp-up. As your tests show, a ramp-up of 5 or so works much better. Even a value 5 means that JMeter has to create 200 sampler threads in one second. That's a lot of work, both for JMeter - and the server and network if JMeter succeeds in doing so. You probably need to use more samples (or repeat the existing ones), so that the first threads are still running when the last threads start. What is the load that your server is expected to handle? On 07/11/2008, Massimiliano Giraldo <[EMAIL PROTECTED]> wrote: > > Hi! > > In my test, I see that the response time is good until 3000-4000 requests > (the total of requests is 10000=1000x10), then some requests (almost 7-8% of > whole test) are slow. > > The structure of the test is: > - Thread Group > - Http Request Defaults > - Http Cookie Manager > - Http Request > - View Result in Table > - Summary Report > > I see this strange behaviour only with 500 users (ramp up = 0) and 1000 > users (ramp up = 0 and 1). > > I show you my results (the meaning of numbers is: Test Number, Users, > Ramp-Up, Average, Deviation, Throughput). > These results is the average between five or more test. > > 1 - 100 - 0 - 18 - 31 - 1343 > 2 - 100 - 1 - 1 - <1 - 945 > 3 - 100 - 10 - 1 - <1 - 100 > ------------------------------------------ > 4 - 500 - 0 - 81,4 - 410 - 1216 > 5 - 500 - 1 - 9 - 24 - 3521 > 6 - 500 - 10 - <1 - <1 - 495 > ------------------------------------------- > 7 - 1000 - 0 - 262 - 887 - 1019 > 8 - 1000 - 1 - 215 - 720 - 1650 > 9 - 1000 - 5 - 1 - 3,4 - 1894 > 10 - 1000 - 10 - 1 - 1 - 976 > ------------------------------------------------------------------- > > As you may notice, in test number 4, 7 and 8, I find these delays. Whereas > in test number 5, 6, 9 and 10, the situation becomes normal. > > Max > > > > > sebb-2-2 wrote: > > > > 10 samples is a very short test; it won't give the systems much time > > to stabilise. > > > > Are you seeing both latency and elapsed time increases? > > What is the relationship between them? > > > > Which HTTP Sampler are you using? > > > > Are you using any timers? > > Does the test plan have lots of listeners and assertions? > > > > Do you get the same symptoms with 100 threads? 500? > > > > Are you sure your network throughput is adequate? > > > > On 07/11/2008, Massimiliano Giraldo <[EMAIL PROTECTED]> wrote: > >> > >> Thanks for the reply, Sipathamandla! > >> > >> I know the meaning of ramp-up period and I know that I'm stressing a lot > >> the > >> system. > >> I'm trying to understand why I have these delays. It seems that, when > >> many > >> threads come together, some of them suffer a delay that seems to be > >> about > >> 700 ms, then 1.5 s, 3 s and then 9 seconds. > >> > >> I'm trying to understand this phenomenon :working: > >> > >> Max > >> > >> > >> > >> > >> Sipathamandla wrote: > >> > > >> > From the manual; > >> > > >> > http://jakarta.apache.org/jmeter/usermanual/component_reference.html#Thread_Group > >> > the ramp up will give you server time to breath by delaying the start > >> > of the next thread, setting it to 0 or 1 means, for zero, the server > >> > is going to get 1000 request instantly and that will definitely impact > >> > response time. Any increase in the ramp up time means the server is > >> > not flooded. > >> > > >> > In the instance of 500 users, 0 seconds ramp up, and a delay of 3 > >> > seconds, you have halved the requests and mathematically i would > >> > expect it to half the response time as well. (might not always be the > >> > case) > >> > > >> > > >> > ____________________________________________________________________________________________ > >> > > >> > > >> > > >> > > >> > 2008/11/7 Massimiliano Giraldo <[EMAIL PROTECTED]>: > >> >> > >> >> Hi all! > >> >> > >> >> I'm testing (with JMeter) Apache Web Server with a cluster of two > >> >> computers > >> >> with JBoss (mod_jk for connection between Web Server and JBoss). > >> >> I make requests to a simple servlet. > >> >> I plan: > >> >> - 1000 users; > >> >> - 0 or 1 second of ramp-up; > >> >> - 10 requests for each user. > >> >> > >> >> I see that the response times sometimes get worse, reaching steps of > >> 1.5 > >> >> seconds, 3 seconds and 9 seconds. > >> >> > >> >> If I plan 5 or 10 seconds of ramp-up, the average response time is > >> about > >> >> 2-3 > >> >> ms (and it's ok!) > >> >> If I plan 500 users and 0 seconds of ramp-up, the delay reaches 3 > >> seconds > >> >> (and it doesn't reach 9 seconds). > >> >> > >> >> I suppose that there are some settings that limit the number of > >> active > >> >> threads (in Jmeter, Apache or JBoss) and therefore, when there are > >> many > >> >> active threads, some of them will be deferred. > >> >> > >> >> Is it correct? > >> >> Was anyone of you in this situation? > >> >> Does anyone know which settings I need to modify to resolve my > >> problem? > >> >> > >> >> Thanks > >> >> > >> >> Max > >> >> > >> >> ----- > >> >> http://www.pcguide.netsons.org http://www.pcguide.netsons.org > >> >> -- > >> >> View this message in context: > >> >> > >> > http://www.nabble.com/Jmeter%2C-Apache-Web-Server-and-JBoss-Cluster%3A-the-response-time-grows%21-tp20376974p20376974.html > >> >> Sent from the JMeter - User mailing list archive at Nabble.com. > >> >> > >> >> > >> >> --------------------------------------------------------------------- > >> >> 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] > >> > > >> > > >> > > >> > >> > >> ----- > >> http://www.pcguide.netsons.org http://www.pcguide.netsons.org > >> -- > >> > >> View this message in context: > >> > http://www.nabble.com/Jmeter%2C-Apache-Web-Server-and-JBoss-Cluster%3A-the-response-time-grows%21-tp20376974p20377763.html > >> > >> Sent from the JMeter - User mailing list archive at Nabble.com. > >> > >> > >> --------------------------------------------------------------------- > >> 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] > > > > > > > > > ----- > http://www.pcguide.netsons.org http://www.pcguide.netsons.org > -- > > View this message in context: > http://www.nabble.com/Jmeter%2C-Apache-Web-Server-and-JBoss-Cluster%3A-the-response-time-grows%21-tp20376974p20380774.html > > Sent from the JMeter - User mailing list archive at Nabble.com. > > > --------------------------------------------------------------------- > 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]
