Re: [Grinder-use] need help in understanding how to analyze the data gathered using grinder
Distributed load testing framework - Java, Jython, or Clojure scripts.
Brought to you by:
philipa
From: Kapil S. <ksa...@gm...> - 2008-12-28 18:45:11
|
Dear Philip, Thanks for your answers so >>It depends on the question - what is a "user"? It's usually convenient >>to think of each test run as a user's browser session. E.g., at the >>beginning of each run the connections are reset and cookies are discarded. the number of concurrent users would be = number of worker threads X number of runs. regarding other questions I am still bit confused though, now going to dive deep into this excellent tool. Regards & thanks again Kapil On Sun, Dec 28, 2008 at 7:48 AM, Philip Aston <ph...@ma...> wrote: > Kapil Sachdeva wrote: > > I am new to the performance testing and saw all the good elements that > > I was looking for in a test tool in grinder. Thank you for writing > > such a great tool. > > > > However, I am not able to really understand some of the terminology > > which is necessary to analyze the data I am gathering by running my > > scripts. The first one > > > > - how do you calculate the "concurrent users" ? > > Assuming only one agent process is it the (number of worker > > threads) or (number of worker threads X test runs). > > It depends on the question - what is a "user"? It's usually convenient > to think of each test run as a user's browser session. E.g., at the > beginning of each run the connections are reset and cookies are discarded. > > > > > - which data in the log files or the console UI will tell me that my > > app supports 'x' number of users at the same time ? or is it that my > > server would start crashing or throwing exception as I increase the > > number of users and I have find to the number at which my server is > > stable and that is the number of concurrent users. > > You need to decide when your server is overloaded. You should probably > set an acceptable target response time, then try with increasing loads > until you reach that response time. > > > > > - If I specify the number of runs in my script I can see that sample > > collection is still happening in console and I can see that the mean > > TPS going down as a result of it. I am not able to understand TPS > > (mean) and TPS (peak) signficance. Are these just some client > > statistics which can be safely ignored ? > > The collection will carry on (and the TPS will continue to fall) until > you tell the console to stop recording. I usually start the worker > processes, wait until the server under test is "warmed up", then start > recording for a fixed number of samples. The console controls allow you > to specify a fixed number of samples after which it will stop recording > automatically. > > The mean TPS is the total number of tests performed divided by the time > since the console started recording. The peak TPS is the highest number > of transactions received in a console sample period, divided by the > sample period - the shorter the console sample period, the more > variability you will see in peak TPS. > > - Phil > > > > ------------------------------------------------------------------------------ > _______________________________________________ > grinder-use mailing list > gri...@li... > https://lists.sourceforge.net/lists/listinfo/grinder-use > |