Also, a ramp-up period of 0 seconds means that JMeter is working much harder, and if running on the same system as the webserver, it will compete for CPU.
There is also some JMeter activity that only runs on the first loop. I'd recommend a ramp-up time that starts between 1 to 10 threads per second - or perhaps more if the box is powerful and/or the test quite simple. The fact that your server coped with 100 threads when started over 80 seconds indicates the problem was most likely the initial peak load - assuming that the test ran long enough for all 100 threads to be running at once at some point. To reduce the effect of startup, the test should run for considerably longer than the startup time - ideally at least 10 times as long. If you want to get some idea of the load JMeter can generate, you can use the Java Test sampler, as this does not rely on external servers or network. If Jmeter struggles to generate a suitable load, then the PC is underpowered (or you have too many listeners). The HTTP Samplers are a bit more complicated than the JavaTest one, and so will use more CPU, but you should be able to get some idea from the test of the load that JMeter can generate. Of course the network will reduce that. S. On 7/7/05, Peter Lin <[EMAIL PROTECTED]> wrote: > if JMeter and the server are both running on the same box, I would say > the results are not valid. 5 concurrent threads is poor, but I would > recommend running the server on a separate system and run the tests > again. > > In "so you want performance" article, I have some basic numbers for > xml performance. compare the results you got with the base results I > have for XML and it will give you a better idea. > > peter > > > On 7/7/05, Andrea Loddo <[EMAIL PROTECTED]> wrote: > > Thanks Peter for you feedback. I have only one question. Before planning > > my tests I' d like to know whether performing tests on my system is > > reasonable and has a sense!I am asking you if a Web Service that > > fails with 5 threads and ramp-up period =0 has a sense. If it has I will > > happy to follow your suggestions and to plan complex tests! > > > > Thanks. > > Andrea > > > > > > Peter Lin wrote: > > > > >welcome to the world of performance and load testing. The short > > >answer to your question is you need to know what the performance > > >requirements are in terms of > > > > > >1. average response time > > >2. max and average concurrent load > > >3. max and average request/sec > > >4. number of machines > > >5. max network bandwidth > > >6. any Service Level Agreements and performance requirements > > > > > > > > >the long answer is you need to know exactly what kind of service your > > >testing and how it will be used. Without that information, stress > > >testing won't tell you much and it won't make your life easier. I > > >would suggest reading my articles to get an overview of what > > >performance means and how to go about designing performance tests. > > > > > >http://jakarta.apache.org/tomcat/articles/performance.pdf > > >http://jakarta.apache.org/tomcat/articles/benchmark_summary.pdf > > >http://cvs.apache.org/~woolfel/article.zip > > > > > >hope that helps > > > > > > > > > > > > > > > > --------------------------------------------------------------------- > > 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] > > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
