jmeter-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Peter Lin <wool...@gmail.com>
Subject Re: jmeter memory consumption
Date Thu, 01 Jul 2004 18:11:23 GMT
With no listeners, you should be able to get decent to good
performance with 500 threads, but I don't see any point doing that.
Running a test without saving the results doesn't tell you anything,
since there's no data proving a server can scale.

I took a quick look at Silk Performer and I don't know if it is java
based. I'm guessing it's C++ based app and isn't actively calculating
statistics, generating graphs and saving the results while the test is
running. You'll have to clue me in on how silk works.

Are you testing 500 threads against a static HTML page, or a real
dynamic page? The reason I ask is 500 concurrent threads without any
ramp up or timer would mean the server is handling 100million+ page
views a day. Back when I worked at a big online directory site, their
peak concurrent requests across the cluster was around 40. It
translated to 1 concurrent request per server average and 2-3
concurrent during peak traffic. It's probably gone up since 2002. The
site used to get over 10million+ page views a day. Keep in mind it's
page views and not requests. The only places I know of with lots of
test machines are big financial companies. I know one of the biggest
financial firms in the US has a couple hundred clients machines for
stress testing.

If you provide more details about the error you're getting in non-gui
mode, I'll gladly take a look to see if it's a real bug. it could be
that non-gui mode is sending requests faster than the server can
handle and is causing requests to hang. You can easily see if that is
happening by looking at the number of open connections to the server.

peter


On Thu, 1 Jul 2004 09:18:50 -0700 (PDT), Remedy QA <remedyqa@yahoo.com> wrote:
> 
> Hi Peter,
> 
> I have been using Segue Silk Performer in the past and was always able to get a load
of 500 on a similar machine described below with good results.  I am actually trying to replace
Silk Performer with JMeter.
> 
> I am trying to see how much load one Jmeter client can handle.  I am planning to use
several clients as you suggested but there are only so many machines available.  My plan is
to get up to a 5000 virtual user load going.  If your approximation of 75 virtual users per
machine is used, that would yield about 67 client machines to generate such a load!
> 
> I suppose no one has used Jmeter to the extent that I am planning.
> 
> mabel
> 
> 
> Peter Lin <woolfel@gmail.com> wrote:
> to my knowledge, even if you use a commercial product like mercury.
> You still can't simulate 250 threads from one system without it eating
> a ton of memory. In fact, I believe mercury doesn't recommend you try
> it, unless you're using a beefy dual or quad CPU server with 8Gb of
> RAM, Gigabit ethernet and Gigabit router.
> 
> in fact, since mercury prefers to save the results to a database,
> you'd have a hard time doing it from one system. Mercury happens to
> have a good reputation and is considered a reliable testing tool.
> 
> I've been able to go up to 75 threads with JMeter with decent
> performance. For more than 100 threads I always use multiple client
> machines.
> 
> peter
> 
> On Wed, 30 Jun 2004 22:29:06 -0400, Michael Stover wrote:
> >
> > So what's the problem, exactly?
> >
> > -Mike
> >
> > On Wed, 2004-06-30 at 20:21, Remedy QA wrote:
> > > It seems Jmeter is a memory hogger. If given more memory, it will keep consuming.
I am using JMeter nightly build of June 12.
> > >
> > > During my test run of approximately 50 minutes, with 250 virtual users on one
GUI Jmeter client, it managed to consume about 1 GB of real memory. As the test continued,
the memory just kept diminishing. The garbage collecting (minor collecting) happened about
every 20 to 60 seconds. The CPU spikes happen when there are GCs.
> > >
> > > I also ran the same test on a machine with only 1 GB of RAM. When the test
was over, real memory was at about 32mb.
> > >
> > > I tried with non-GUI mode but several threads hung and never was able to finish.
> > >
> > > So it seems that if I use a machine with more memory and give it a bigger heap,
it just consumes as much as it can. I don't think 250 virtual users for the machine type I
use is too much load. There must be something I am missing. Any help appreciated.
> > >
> > > I ran a test with the following configuration:
> > >
> > > Single JMeter Client on Windows 2000 Server, 2 GB RAM, single 2.8 Ghz Pentium
4 CPU. JDK 1.4.2_04
> > >
> > > JMeter JVM settings:
> > > set HEAP=-Xms1280m -Xmx1280m
> > > set NEW=-XX:NewSize=512m -XX:MaxNewSize=512m
> > > set DEBUG=-verbose:gc -XX:-PrintTenuringDistribution -XX:+PrintGCDetails -XX:+PrintGCTimeStamps
-XX:+PrintGCApplicationConcurrentTime -XX:+PrintGCApplicationStoppedTime
> > >
> > > All other JVM settings are the defaults that came with jmeter.bat.
> > >
> > > JMeter output set to CSV, jmeter and jorphan logging set to warning.
> > >
> > > Jmeter script:
> > >
> > > Test Plan
> > > ThreadGroup - 250 virtual users, ramping up every 5 seconds. Loop once.
> > > Aggregate Listener
> > > Simple Controller
> > > 8 HTTP Requests in here
> > > 3 Aggregate Listeners in here
> > > Runtime Controller - 45 minutes total for all users.
> > > 33 HTTP Requests in here
> > > 2 Aggregate Listeners in here
> > > Simple Controller
> > > 2 HTTP Requests in here.
> > >
> > >
> > >
> > >
> > > ---------------------------------
> > > Do you Yahoo!?
> > > Yahoo! Mail is new and improved - Check it out!
> > --
> > Michael Stover
> > Apache Software Foundation
> >
> > ---------------------------------------------------------------------
> > 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
> 
>                
> ---------------------------------
> Do you Yahoo!?
> New and Improved Yahoo! Mail - 100MB free storage!

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