jmeter-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Peter Lin <wool...@gmail.com>
Subject Re: Memory leaks - large and uncollected
Date Wed, 03 Aug 2005 20:25:04 GMT
glad that fixed your problem. like mike said, the only way is find & replace.

Not to excuse Sun for a lame SSL implementation, but most sites using
SSL typically use hardware acceleartion.  from a price/performance
perspective, using hardware SSL accelerator is much more manageable.
SSL performance is pretty linear and will consume copious amounts of
CPU cycles and RAM, so it can lead to instability of the server. From
past experience with apache and IIS, using software SSL can lead to
server instability.


peter


On 8/3/05, Jon Andersen <janderse@umich.edu> wrote:
> Peter,
> 
> It appears that Sun SSL -:> OutOfMemoryException is the culprit.  I ran
> the same script without SSL, and don't see the memory usage growing.
> 
> Is there some way to switch from the default "HTTP Request"
> implementation to the "HTTP Request HttpClient" implementation without
> rewriting the script?  Do both use the same configuration, so I could
> do a search-n-replace?
> 
> Thanks a bunch!
> 
> -Jon Andersen
> Software developer
> 734-260-6083 (work)
> 734-646-5577 (home)
> Digital Media Commons - Duderstadt Center
> University of Michigan
> 
> On Aug 3, 2005, at 3:14 PM, Peter Lin wrote:
> 
> > I've run optimizeIt on Jmeter about 3 dozen times when I wrote the
> > various samplers like Webservice, jms, accesslog and tomcat monitor.
> > Under normal conditions, JMeter's memory usage is constant. I've run
> > tests using OptimizeIt on JMeter for 500K, 1million and 2 million
> > requests.
> >
> > what I haven't done is try it on SSL. I believe there's a known bug
> > with Sun's SSL implementation which results in connections not closing
> > and resulting in OutOfMemoryException.  The solution to that is to use
> > HttpClient sampler instead of the default one.
> >
> > the default http sampler uses the Sun implementation, which I believe
> > exhibits that behavior. the HttpClient version should perform better.
> > You can also verify this by using HTTP with the same test instead of
> > HTTPS. If the bug doesn't appear in non-secure HTTP, it would support
> > the idea that HTTPS is the cause.
> >
> > I know that some EJB containers use Sun implementation for SSL/HTTPS
> > and they exhibit the same exact bug. If those connections are
> > time_wait and aren't getting cleaned up, it's very likely the known
> > SSL bug.
> >
> > peter
> >
> >
> >
> > On 8/3/05, janderse@umich.edu <janderse@umich.edu> wrote:
> >> Hi,
> >>
> >> I've created a script which generates thousands of HTTP requests; the
> >> script seems to work fine, but I'm seeing a large memory leak in
> >> JMeter
> >> itself.  Before you write this off as a "newbie" question, here is
> >> everything I've tried to do to fix it:
> >>
> >> 1.  Deleting _all_ listeners.  I still see the memory leak, even when
> >> running without any listeners.
> >> 2.  Stopping the test and doing "Clear all".  The memory is _not_
> >> reclaimed, even after a full garbage collection.
> >> 3.  Increasing the memory available to JMeter to 1GB.  As the test
> >> runs,
> >> the memory usage climbs until its doing continuous full garbage
> >> collections.  Eventually an OutOfMemory exception occurs.
> >> 4.  Switching the HTML parser from the default to the
> >> RegexpHTMLParser.
> >> Still see the same memory leak, albeit slower.
> >>
> >> Does this sound familiar to anyone?  It appears to be a slow leak
> >> per-request.  I'm running XP against an SSL-enabled server.  Does SSL
> >> leak
> >> memory per-connection?  "netstat -n" shows thousands of connections -
> >> and
> >> it looks like they don't go away until JMeter quits.
> >>
> >> I only see the leak if I do on the order of 100,000+ HTTP requests.
> >> I can
> >> send interested parties the script to test for themselves.  I'm not
> >> attaching the script now since its 500KB.
> >>
> >> I really need to figure out what is going wrong.  I'm about to
> >> attempt to
> >> run a memory profiler (OptimizeIt) on JMeter to determine where the
> >> leak is
> >> under-the-covers...
> >>
> >> Thanks,
> >>
> >> -Jon Andersen
> >> The Sakai Project
> >> http://www.sakaiproject.org
> >>
> >> ---------------------------------------------------------------------
> >> To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
> >> For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
> >>
> >>
> >
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
> > For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
> >
> >
> >
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
> 
>

---------------------------------------------------------------------
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