jmeter-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From git <>
Subject Re: Variance between JMeter response times and real users
Date Tue, 16 Jan 2007 17:13:37 GMT

Other than this being because your script is missing stuff (people often
forget to include all the image downloads, css, javascript etc) there
are other reasons that I have seen this happen.

1) It is quite hard to get JMeter not to dispatch the requests to the
server in a 'nice' way, i.e. as a continuous stream.  Real requests tend
to come in chunks, lots of requests then a gap.  Some servers perform
very much worse under these 'chunky' conditions than under smooth load.

2) Users get annoyed at waiting and do stuff like hitting submit more
than once.  That pushes up load, and so performance falls and so they do
it more.  This positive feedback from the users causes a downward spiral
in performance.

If it looks like the problem is number 1 - I would look first at the DB.
The bottle neck is often in the way transactions are serialised on the

If it looks like the problem is number 2 - I might suggest using
Javascript to prevent double submission, but that is a lot of work.



On Tue, 2007-01-16 at 07:56 -0800, rgjohnson wrote:

> We have used JMeter to perform load testing on a webserver. One major problem
> we have encountered is a large variance between the load JMeter says the
> server can handle vs. the load the server can handle when real users are
> accessing the application. In one example, we used JMeter to simulate 60
> users running on a single instance of the webserver running an order entry
> application. All response times and statistics were very favorable. We then
> put real users on the same order entry appllication and started running into
> unaccetable performace and response times after about 15 users. I can think
> of several reasons why response times reported by JMeter could be worse than
> real life, but I am having a difficult time understanding how response times
> and server performance can be so much better when simulating users with
> JMeter than what the response times and server performance are with real
> users.  Has anyone else had this kind of experience?

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message