well, it shouldn't stay exactly the same, but it could be mathematically possible to get a similar throughtput, though very unlikely. so it matters also what timers you use, how fast the server responds to these requests etc etc.
depending on your test plan and the timers you set, it is possible that the throughtput will not increase over a specific value, even if you don't use a constant throughput timer. answer to q4. it depends on specifications. but when lacking these, set a convenient target for the max / 90 % line / median response times. set a realistic workload for the application and target usres. THEN you can see for what number of threads these numbers are exceeded! that will give you a good answer to what is the limit of users and number of requests that the server can handle the way you would wish. this is the minimum you have to do. you should consider having a realistic test plan for the specific situation you benchmark the application, isolated test environment and good results from a statistical perspective etc. http://wiki.apache.org/jakarta-jmeter/FrontPage http://www.perftestplus.com/pubs.htm (some good links on performance testing, this will answer some generic questions) one example: one customer requested that the max is not higher than 3s on the hardware we recommended. at the client's test site, I tested with 100 users (the number of concurrent users of their license) and with 10 times more requests than a normal-use scenario. got response times with a max of 0.5s. everybody was happy. we know the application can handle more users, but we had to prove it worked well for them (only 100 users for their license), so the test simulated that every user would click on something every couple of seconds or so all through the test (absurd, but they liked hearing that even so the server answers almost instantly). On Mon, Sep 7, 2009 at 2:44 PM, Thomas G. Lau<[email protected]> wrote: > Dear all, > > just wondering how do you interpret aggregate report result, currently > testing 3 page of web site: > > 1. why number of threads increase, number of sample will increase? what is > the relationship between number of sample with rampup period + number of > threads? > 2. I got average throughput is 3.5/sec, it stay the same on 300, 400,500 > threads with 300 rampup second. Number of sample increase, but average > latency increase, isn't latency increase will driven throughput decrease? > 3. what is the meaning of average latency? from request that sent from > jmeter until server response back 200 ok? > 4. if I want to measure a web server max capacity, how should I do that? > > Thank you. > > > --------------------------------------------------------------------- > 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]
