jmeter-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Michael Stover <mstov...@apache.org>
Subject RE: Hanging threads in non-GUI mode
Date Thu, 01 Jul 2004 21:15:55 GMT
On Thu, 2004-07-01 at 15:57, Remedy QA wrote:
> Hmm, didn't know there were two options. Looking at the June 12th nightly build I'm using,
the Sampler gives "HTTP Request" and "HTTPRequest HTTP Client(ALPHA)" . I am using the default
http sampler because that is what the Proxy recorder uses.  I have been using this since version
1.9.  
>  
> All requests use keep alive.  Just one or two use redirect, but that is only done once
per user (done for login).
>  
> To answer Mike's questions:
>  
> 1. Are you using any timers? 250 virtual users with no delays is a hell
> of a load
> Yes. Between controllers (there are 2 that are repeated for 45 minutes), I have uniform
random timer between 5 to 15 seconds.  Then I have some "user" think time between opening
menus that are uniform random between 1.5 to 5 seconds.  And some uniform random timers between
1 to 4 seconds for "waiting" to click on a link.  Then there are some requests that are performed
without timers.
>  
> 2.  What kind of average response times are you getting from these
> requests?
> The overall average total is 33 seconds. Some pages have it set to retrieve embedded
resources.

This could be a significant source of memory use, to parse the page and
retrieve certain elements.  I find better solutions to be:
1)Using the proxy server, allow it to record all static requests and
keep them in the test plan.  Allows you to turn off redirect-following
and retrieving embedded resources.
2)Figure out the approximate production ratio of page request/static
resource download and make explicit requests for the static elements of
your site, put them inside a throughput controller.  Since browsers
usually cache these items, these resources are downloaded somewhat less
frequently than the dynamic pages, so this helps reproduce that effect.

Other than this, your test plan doesn't sound all that stressful for
JMeter.

> 
>  
> 3. What order of magnitude are the responses? 10's of bytes? 100's?
> 1000's? 10's of KB?
> The average request response size is 42KB.  The minimum response time is less than 1KB

> The largest response is 394KB which happens when opening a page.
>  
> 4. What kind of dynamic elements do you have in the test plan? Are you
> doing a lot of regex parsing of response data?
> Out of 44 HTTP Requests, I have 7 regex parsing and 14 Assertions
> There are 6 aggregate listeners. Running without the listeners did not make a difference.

What difference are you looking for, exactly?  No matter what you do,
JMeter is going to take up whatever memory you lay out for it.  But, I
still don't understand why that's a problem.

-Mike

> 
> "BAZLEY, Sebastian" <sebastian.bazley@atosorigin.com> wrote:
> 
> Another thing - are you using the default HTTP sampler (uses the JVM http
> implementation) or the new sampler (uses Apache HTTPClient)?
> 
> Also, keep-alive? Follow redirects etc?
> 
> S.
> -----Original Message-----
> From: Michael Stover [mailto:mstover1@apache.org]
> Sent: 01 July 2004 19:04
> To: JMeter Users List
> Subject: Re: Hanging threads in non-GUI mode
> 
> 
> Yes, or at least some other things I'd like to know about the test:
> 1. Are you using any timers? 250 virtual users with no delays is a hell
> of a load
> 2. What kind of average response times are you getting from these
> requests?
> 3. What order of magnitude are the responses? 10's of bytes? 100's?
> 1000's? 10's of KB?
> 4. What kind of dynamic elements do you have in the test plan? Are you
> doing a lot of regex parsing of response data?
> 
> -Mike
> 
> On Thu, 2004-07-01 at 13:49, BAZLEY, Sebastian wrote:
> > It would be useful to be able investigate and fix this; I don't suppose
> that
> > you have a reproducible test case that you can post to Bugzilla?
> > 
> > We use batch mode almost exclusively, and apart from a concurrent
> > modification bug that was found and fixed several months ago, it works
> fine.
> > 
> > S.
> > -----Original Message-----
> > From: Remedy QA [mailto:remedyqa@yahoo.com]
> > Sent: 01 July 2004 17:27
> > To: JMeter Users List
> > Subject: RE: jmeter memory consumption
> > 
> > 
> > Well, I did try non-gui mode but some threads kept hanging. It didn't
> happen
> > with gui mode though.
> > 
> > 
> > 
> >
> 
> 		
> ---------------------------------
> Do you Yahoo!?
> Yahoo! Mail - 50x more storage than other providers!
-- 
Michael Stover <mstover1@apache.org>
Apache Software Foundation


---------------------------------------------------------------------
To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jmeter-user-help@jakarta.apache.org


Mime
View raw message