You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@jmeter.apache.org by Cristian Opincaru <cr...@gmail.com> on 2007/06/25 09:55:24 UTC
Throughput Measurement / Tomcat / AXIS
Hi all,
I'm trying to determine the throughput for some webservices implemented in
AXIS 1.4 which is running on top of Tomcat. The services are hosted on a
Linux Pentium 4 machine; I'm using another similar machine to run JMeter.
The puzzling thing is that I'm not able to bring the processor on my server
to 100%. It goes to about 90%, but not higher. Since the processor is 10%
idle, I assume that the throughput measurements are incorrect.
I'm trying to figure out why the cpu does not get to 100% usage. It's not
the network (the two are connected by a 100MB link). The simplest service is
a hello world, where no disk access is involved, therefore, I assume that
the throughput bottleneck should be the processor. But it's not 100% used.
I would appreciate any help!
Thank you!
Cristian
--
Cristian OPINCARU
University of the Federal Armed Forces Munich
http://www.unibw.de/cristian.opincaru
Re: Throughput Measurement / Tomcat / AXIS
Posted by sebb <se...@gmail.com>.
On 25/06/07, Cristian Opincaru <cr...@gmail.com> wrote:
> Please forward me the link about Google testing (I could not find it on the
> list).
>
> I read an IBM paper about Performance Testing for an ESB (
> http://www.research.ibm.com/trl/people/mich/pub/200609_icws2006esbperf.pdf),
> and they said that in order to determine the throughput the CPU utilization
> should be 100%. It makes sense since the CPU is the component that is most
> solicited - my Web services were not writing on the disk and I allocated
> plenty of memory to Tomcat 1GB. I don't know about the sockets, is there a
> way to configure these things?
There should be a way of configuring the maximum sockets - check the
OS documentation.
Also the Tomcat config may need tweaking to allocate enough sessions etc.
> The client machine is only running 25% CPU. So the bottleneck must be the
> server.
Probably, but not necessarily - there could be some other resource
starvation, or indeed some hidden wait in the JMeter test.
> Thankx for the answer!
> Cristian
>
> On 6/25/07, sebb <se...@gmail.com> wrote:
> >
> > On 25/06/07, Cristian Opincaru <cr...@gmail.com> wrote:
> > > Hi all,
> > >
> > > I'm trying to determine the throughput for some webservices implemented
> > in
> > > AXIS 1.4 which is running on top of Tomcat. The services are hosted on a
> > > Linux Pentium 4 machine; I'm using another similar machine to run
> > JMeter.
> > >
> > > The puzzling thing is that I'm not able to bring the processor on my
> > server
> > > to 100%. It goes to about 90%, but not higher. Since the processor is
> > 10%
> > > idle, I assume that the throughput measurements are incorrect.
> >
> > Why? There may be resource waits (e.g. socket closes) involved.
> >
> > Also, do you really want to run your server at 90% CPU? I don't know
> > if you watched the Google talk about performance testing (link posted
> > to this mailing list recently), but the presenter said that one should
> > aim for at most 80% busy on Unix.
> >
> > > I'm trying to figure out why the cpu does not get to 100% usage. It's
> > not
> > > the network (the two are connected by a 100MB link). The simplest
> > service is
> > > a hello world, where no disk access is involved, therefore, I assume
> > that
> > > the throughput bottleneck should be the processor. But it's not 100%
> > used.
> >
> > What is the resource utilisation on the JMeter machine?
> >
> > What are the other resource utilisations (memory, disk)?
> >
> > If the JMeter system is running short of resources, you can reduce the
> > JMeter resource requirements by using non-GUI mode, using only one
> > listener, and reducing Assertions as far as possible.
> >
> > Or you can add a second JMeter machine.
> >
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
> > For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
> >
> >
>
>
> --
> Cristian OPINCARU
> University of the Federal Armed Forces Munich
> http://www.unibw.de/cristian.opincaru
>
---------------------------------------------------------------------
To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
Re: Throughput Measurement / Tomcat / AXIS
Posted by Cristian Opincaru <cr...@gmail.com>.
Please forward me the link about Google testing (I could not find it on the
list).
I read an IBM paper about Performance Testing for an ESB (
http://www.research.ibm.com/trl/people/mich/pub/200609_icws2006esbperf.pdf),
and they said that in order to determine the throughput the CPU utilization
should be 100%. It makes sense since the CPU is the component that is most
solicited - my Web services were not writing on the disk and I allocated
plenty of memory to Tomcat 1GB. I don't know about the sockets, is there a
way to configure these things?
The client machine is only running 25% CPU. So the bottleneck must be the
server.
Thankx for the answer!
Cristian
On 6/25/07, sebb <se...@gmail.com> wrote:
>
> On 25/06/07, Cristian Opincaru <cr...@gmail.com> wrote:
> > Hi all,
> >
> > I'm trying to determine the throughput for some webservices implemented
> in
> > AXIS 1.4 which is running on top of Tomcat. The services are hosted on a
> > Linux Pentium 4 machine; I'm using another similar machine to run
> JMeter.
> >
> > The puzzling thing is that I'm not able to bring the processor on my
> server
> > to 100%. It goes to about 90%, but not higher. Since the processor is
> 10%
> > idle, I assume that the throughput measurements are incorrect.
>
> Why? There may be resource waits (e.g. socket closes) involved.
>
> Also, do you really want to run your server at 90% CPU? I don't know
> if you watched the Google talk about performance testing (link posted
> to this mailing list recently), but the presenter said that one should
> aim for at most 80% busy on Unix.
>
> > I'm trying to figure out why the cpu does not get to 100% usage. It's
> not
> > the network (the two are connected by a 100MB link). The simplest
> service is
> > a hello world, where no disk access is involved, therefore, I assume
> that
> > the throughput bottleneck should be the processor. But it's not 100%
> used.
>
> What is the resource utilisation on the JMeter machine?
>
> What are the other resource utilisations (memory, disk)?
>
> If the JMeter system is running short of resources, you can reduce the
> JMeter resource requirements by using non-GUI mode, using only one
> listener, and reducing Assertions as far as possible.
>
> Or you can add a second JMeter machine.
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
>
>
--
Cristian OPINCARU
University of the Federal Armed Forces Munich
http://www.unibw.de/cristian.opincaru
Re: Throughput Measurement / Tomcat / AXIS
Posted by sebb <se...@gmail.com>.
On 25/06/07, Cristian Opincaru <cr...@gmail.com> wrote:
> Hi all,
>
> I'm trying to determine the throughput for some webservices implemented in
> AXIS 1.4 which is running on top of Tomcat. The services are hosted on a
> Linux Pentium 4 machine; I'm using another similar machine to run JMeter.
>
> The puzzling thing is that I'm not able to bring the processor on my server
> to 100%. It goes to about 90%, but not higher. Since the processor is 10%
> idle, I assume that the throughput measurements are incorrect.
Why? There may be resource waits (e.g. socket closes) involved.
Also, do you really want to run your server at 90% CPU? I don't know
if you watched the Google talk about performance testing (link posted
to this mailing list recently), but the presenter said that one should
aim for at most 80% busy on Unix.
> I'm trying to figure out why the cpu does not get to 100% usage. It's not
> the network (the two are connected by a 100MB link). The simplest service is
> a hello world, where no disk access is involved, therefore, I assume that
> the throughput bottleneck should be the processor. But it's not 100% used.
What is the resource utilisation on the JMeter machine?
What are the other resource utilisations (memory, disk)?
If the JMeter system is running short of resources, you can reduce the
JMeter resource requirements by using non-GUI mode, using only one
listener, and reducing Assertions as far as possible.
Or you can add a second JMeter machine.
---------------------------------------------------------------------
To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jmeter-user-help@jakarta.apache.org