Re: Distributed test, scalability!

[Peter Lin <wo...@gmail.com>]

perhaps mike or sebb can answer that question. I haven't looked at the
remote code close enough to know.

peter



On 9/28/05, Krog Jensen Lars <La...@six.se> wrote:
>
> Hello,
>
> This is indeed one solution, but the problem is that we would need
> buy a lot more NIC and install them into all machines. Another solution
> could be that the controller <-> load agent communication should be
> statistical. What I mean is the agent aggregates a summary and then
> sends it back at constant rate; let's say every 5 seconds or other user
> determined value.
>
> This would be especially useful when using an Aggregate listener, which
> does not display each request invocation.
>
> I downloaded the source code, but I have not figured out yet where to
> start look.
>
> You wouldn't know how the Aggregate listener works when running multiple
> load agents?
>
> Regards
> Lars
>
>
>
> -----Original Message-----
> From: Peter Lin [mailto:woolfel@gmail.com]
> Sent: den 28 september 2005 15:47
> To: JMeter Users List
> Subject: Re: Distributed test, scalability!
>
> what you observed is a fundamental limitation of stress testing. the
> only
> good way to deal with this is to have 2 ethernet ports on each client
> machine and use 2 routers.
>
> this way, all the traffic to the server is on one router and all other
> traffic is on the other router. this is how I have my home environment
> setup.
>
>
> peter
>
>
> On 9/28/05, Krog Jensen Lars <La...@six.se> wrote:
> >
> > Hello,
> >
> >
> >
> > Being new to JMeter I have some questions regarding the scalability of
> > JMeter.
> >
> >
> >
> > I am planning to load test with 10 client machines, but I am running
> in
> > to scalability
> >
> > issues when using 2 load severs.
> >
> >
> >
> > When my test plan is running on the controller host it will generate
> 25
> > Mbps in network traffic
> >
> > (seen from the Win XP Task Mgr Network tab), but when I remote start a
> > load server, the traffic is increased
> >
> > to 65 Mbps. From what I can tell these 40 Mbps is JMeter internal
> > traffic.
> >
> >
> >
> > By disabling all listeners this traffic is totally removed, but even a
> > simple listener as Simple Data Writer is
> >
> > generating 40 Mbps traffic.
> >
> > It is pretty useless running in without listeners.
> >
> >
> >
> > Another related question is whether the Aggregate Report listener
> really
> > aggregate the result from all the slaves?
> >
> > The reason for asking is that when I start a testplan from the
> > controller I have a throughput of ~100 pages /sec, but when
> >
> > I remote start a load server, I would expect to receive ~200 pages
> /sec.
> >
> >
> > If I start JMeter locally on two client hosts each will receive ~ 100
> > pages / sec.
> >
> >
> >
> >
> >
> > Besides that, JMeter is really, really nice and useful.
> >
> >
> >
> > Best Regards
> >
> > Lars Krog-Jensen
> >
> > SIX AB
> >
> > Stockholm
> >
> >
> >
>

Re: Distributed test, scalability!

[sebb <se...@gmail.com>]

If you run all the tests independently using non-GUI mode, you'll
probably find that fewer test boxes are needed (unless the limiting
factor is network bandwidth).

Adding a summary collector is useful for non-GUI runs (see jmeter.properties).

When the tests finish, the test results can be combined and analysed -
or analysed and then combined.

==

The code is in the various Remote*.java files.

Dunno whether it would be possible to pass back summary information -
there would need to be a test element that accepted it.

S.
On 28/09/05, Peter Lin <wo...@gmail.com> wrote:
> perhaps mike or sebb can answer that question. I haven't looked at the
> remote code close enough to know.
>
> peter
>
>
>
> On 9/28/05, Krog Jensen Lars <La...@six.se> wrote:
> >
> > Hello,
> >
> > This is indeed one solution, but the problem is that we would need
> > buy a lot more NIC and install them into all machines. Another solution
> > could be that the controller <-> load agent communication should be
> > statistical. What I mean is the agent aggregates a summary and then
> > sends it back at constant rate; let's say every 5 seconds or other user
> > determined value.
> >
> > This would be especially useful when using an Aggregate listener, which
> > does not display each request invocation.
> >
> > I downloaded the source code, but I have not figured out yet where to
> > start look.
> >
> > You wouldn't know how the Aggregate listener works when running multiple
> > load agents?
> >
> > Regards
> > Lars
> >
> >
> >
> > -----Original Message-----
> > From: Peter Lin [mailto:woolfel@gmail.com]
> > Sent: den 28 september 2005 15:47
> > To: JMeter Users List
> > Subject: Re: Distributed test, scalability!
> >
> > what you observed is a fundamental limitation of stress testing. the
> > only
> > good way to deal with this is to have 2 ethernet ports on each client
> > machine and use 2 routers.
> >
> > this way, all the traffic to the server is on one router and all other
> > traffic is on the other router. this is how I have my home environment
> > setup.
> >
> >
> > peter
> >
> >
> > On 9/28/05, Krog Jensen Lars <La...@six.se> wrote:
> > >
> > > Hello,
> > >
> > >
> > >
> > > Being new to JMeter I have some questions regarding the scalability of
> > > JMeter.
> > >
> > >
> > >
> > > I am planning to load test with 10 client machines, but I am running
> > in
> > > to scalability
> > >
> > > issues when using 2 load severs.
> > >
> > >
> > >
> > > When my test plan is running on the controller host it will generate
> > 25
> > > Mbps in network traffic
> > >
> > > (seen from the Win XP Task Mgr Network tab), but when I remote start a
> > > load server, the traffic is increased
> > >
> > > to 65 Mbps. From what I can tell these 40 Mbps is JMeter internal
> > > traffic.
> > >
> > >
> > >
> > > By disabling all listeners this traffic is totally removed, but even a
> > > simple listener as Simple Data Writer is
> > >
> > > generating 40 Mbps traffic.
> > >
> > > It is pretty useless running in without listeners.
> > >
> > >
> > >
> > > Another related question is whether the Aggregate Report listener
> > really
> > > aggregate the result from all the slaves?
> > >
> > > The reason for asking is that when I start a testplan from the
> > > controller I have a throughput of ~100 pages /sec, but when
> > >
> > > I remote start a load server, I would expect to receive ~200 pages
> > /sec.
> > >
> > >
> > > If I start JMeter locally on two client hosts each will receive ~ 100
> > > pages / sec.
> > >
> > >
> > >
> > >
> > >
> > > Besides that, JMeter is really, really nice and useful.
> > >
> > >
> > >
> > > Best Regards
> > >
> > > Lars Krog-Jensen
> > >
> > > SIX AB
> > >
> > > Stockholm
> > >
> > >
> > >
> >
>
>

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