You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@jmeter.apache.org by alwaysbmore <al...@yahoo.com> on 2010/12/27 15:51:01 UTC
Multiple Jmeter Instances
I get two different totally different results. When running multiple JMeters
instances on multiples machines compared to a single Jmeter instance on a
single machine. For example, If I'm looking to run 100 concurrent users. If
I run all 100 threads from a single machine the results are much lower, if I
were to run 50 threads on one machine and 50 on another. Both instances are
run separate from each other, not as a master and server.
I suspect, it might have something to do with all my how the server is
configured to accept request all coming from a single IP Address.
Can anyone one confirm this, and give some feedback on what results should I
trust?
--
View this message in context: http://jmeter.512774.n5.nabble.com/Multiple-Jmeter-Instances-tp3319388p3319388.html
Sent from the JMeter - User mailing list archive at Nabble.com.
---------------------------------------------------------------------
To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
Re: Multiple Jmeter Instances
Posted by Anthony Johnson <an...@gmail.com>.
The SimpleDataWriter could be your bottleneck as the test scales. For
each sample an entry needs to be written to an output file. So as
your test scales, so does the dependency on the speed of your
hard-drive. This could also be compounded by the fact that you are
running in a VM which will have slower hard drive reads and writes and
may be missing some filesystem optimizations.
I would still suggest disabling the SimpleDataWriter and observing the effect.
Thanks,
Anthony
On Mon, Dec 27, 2010 at 10:48 AM, alwaysbmore <al...@yahoo.com> wrote:
>
> Well I was using 100 just as an example. 1000 is more realistic. So 1000 on
> a single instance compared to 500 split between two instances. I'm running
> JMeter 2.4. on Windows Server 2003 which is a virtual machine. I do not
> have any listeners expect for a Simple Data Writer.
>
> What I'm having hard time understanding is just the difference between
> running on a single VM and multiple VMs. If I was to run 1000 threads on a
> single Jmeter instance on one VM, my test runs fine, much lower response
> times and no errors. If I run two VM's both at 500 threads exact same test
> plan. My response times are a lot higher and I receive a lot more errors.
> They are mostly connection timeout responses.
> --
> View this message in context: http://jmeter.512774.n5.nabble.com/Multiple-Jmeter-Instances-tp3319388p3319436.html
> Sent from the JMeter - User mailing list archive at Nabble.com.
>
> ---------------------------------------------------------------------
> 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
Re: Multiple Jmeter Instances
Posted by Felix Frank <ff...@mpexnet.de>.
On 12/27/2010 04:48 PM, alwaysbmore wrote:
>
> Well I was using 100 just as an example. 1000 is more realistic. So 1000 on
> a single instance compared to 500 split between two instances. I'm running
> JMeter 2.4. on Windows Server 2003 which is a virtual machine. I do not
> have any listeners expect for a Simple Data Writer.
>
> What I'm having hard time understanding is just the difference between
> running on a single VM and multiple VMs. If I was to run 1000 threads on a
> single Jmeter instance on one VM, my test runs fine, much lower response
> times and no errors. If I run two VM's both at 500 threads exact same test
> plan. My response times are a lot higher and I receive a lot more errors.
> They are mostly connection timeout responses.
Are you using constant throughput timers per chance?
Otherwise, your Jmeter processes may be limited by several factors,
among them being memory per VM (more garbage collection when low on
memory) and possibly thread concurrency?
Regards,
Felix
---------------------------------------------------------------------
To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
Re: Multiple Jmeter Instances
Posted by Felix Frank <ff...@mpexnet.de>.
On 07/14/2011 01:09 PM, SD wrote:
> Please let me know how to set up VMs environment, in my case I have these VMs
> outside of the country, from here and want to run Jmeter performance scripts
> in VMs. Help me out as I am new to this.
Hi,
setting up Jmeter in a VM is no different than doing so in a physical
environment.
If you end up facing any specific issues, come back and ask for help.
Please start a new thread if and when you do so.
Regards,
Felix
---------------------------------------------------------------------
To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
Re: Multiple Jmeter Instances
Posted by SD <dr...@yahoo.com>.
alwaysbmore <alwysbmore <at> yahoo.com> writes:
>
>
> Well I was using 100 just as an example. 1000 is more realistic. So 1000 on
> a single instance compared to 500 split between two instances. I'm running
> JMeter 2.4. on Windows Server 2003 which is a virtual machine. I do not
> have any listeners expect for a Simple Data Writer.
>
> What I'm having hard time understanding is just the difference between
> running on a single VM and multiple VMs. If I was to run 1000 threads on a
> single Jmeter instance on one VM, my test runs fine, much lower response
> times and no errors. If I run two VM's both at 500 threads exact same test
> plan. My response times are a lot higher and I receive a lot more errors.
> They are mostly connection timeout responses.
Please let me know how to set up VMs environment, in my case I have these VMs
outside of the country, from here and want to run Jmeter performance scripts
in VMs. Help me out as I am new to this.
---------------------------------------------------------------------
To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
Re: Multiple Jmeter Instances
Posted by alwaysbmore <al...@yahoo.com>.
Well I was using 100 just as an example. 1000 is more realistic. So 1000 on
a single instance compared to 500 split between two instances. I'm running
JMeter 2.4. on Windows Server 2003 which is a virtual machine. I do not
have any listeners expect for a Simple Data Writer.
What I'm having hard time understanding is just the difference between
running on a single VM and multiple VMs. If I was to run 1000 threads on a
single Jmeter instance on one VM, my test runs fine, much lower response
times and no errors. If I run two VM's both at 500 threads exact same test
plan. My response times are a lot higher and I receive a lot more errors.
They are mostly connection timeout responses.
--
View this message in context: http://jmeter.512774.n5.nabble.com/Multiple-Jmeter-Instances-tp3319388p3319436.html
Sent from the JMeter - User mailing list archive at Nabble.com.
---------------------------------------------------------------------
To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
Re: Multiple Jmeter Instances
Posted by Felix Frank <ff...@mpexnet.de>.
> @Felix No, I'm not using constant through put timer, should I?
No :)
Not if you want to distribute the testing.
---------------------------------------------------------------------
To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
Re: Multiple Jmeter Instances
Posted by Adrian Speteanu <as...@gmail.com>.
Hmmm, running a test with a large number of threads from a single virtual
machine is also tricky, I wouldn't recommend it, unless you have no
alternatives. The overhead from the virtual machine might slow you down from
time to time, though not necessarily. However, when you run into problems
such as this you never know where to start looking for the cause.
On Mon, Dec 27, 2010 at 10:14 PM, Deepak Shetty <sh...@gmail.com> wrote:
> and two-four cpu's? I doubt you can accurately simulate 1000 threads unless
> you have reasonable think times in your test. Check with perfmon , it
> should
> tell you that your Jmeter client is overloaded.
>
> regards
> deepak
>
> On Mon, Dec 27, 2010 at 12:11 PM, alwaysbmore <al...@yahoo.com>
> wrote:
>
> >
> > This is a:
> > Windows Server 2003
> > 2.53 GHz 3.0 GB of Ram
> > Virtual Machine running on VMware.
> >
> > I also increased the amount of memory allocated to the JVM.
> > set HEAP=-Xms1024m -Xmx1024m
> >
> >
> >
> > --
> > View this message in context:
> >
> http://jmeter.512774.n5.nabble.com/Multiple-Jmeter-Instances-tp3319388p3319678.html
> > Sent from the JMeter - User mailing list archive at Nabble.com.
> >
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
> > For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
> >
> >
>
Re: Multiple Jmeter Instances
Posted by Deepak Shetty <sh...@gmail.com>.
and two-four cpu's? I doubt you can accurately simulate 1000 threads unless
you have reasonable think times in your test. Check with perfmon , it should
tell you that your Jmeter client is overloaded.
regards
deepak
On Mon, Dec 27, 2010 at 12:11 PM, alwaysbmore <al...@yahoo.com> wrote:
>
> This is a:
> Windows Server 2003
> 2.53 GHz 3.0 GB of Ram
> Virtual Machine running on VMware.
>
> I also increased the amount of memory allocated to the JVM.
> set HEAP=-Xms1024m -Xmx1024m
>
>
>
> --
> View this message in context:
> http://jmeter.512774.n5.nabble.com/Multiple-Jmeter-Instances-tp3319388p3319678.html
> Sent from the JMeter - User mailing list archive at Nabble.com.
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
>
>
Re: Multiple Jmeter Instances
Posted by alwaysbmore <al...@yahoo.com>.
This is a:
Windows Server 2003
2.53 GHz 3.0 GB of Ram
Virtual Machine running on VMware.
I also increased the amount of memory allocated to the JVM.
set HEAP=-Xms1024m -Xmx1024m
--
View this message in context: http://jmeter.512774.n5.nabble.com/Multiple-Jmeter-Instances-tp3319388p3319678.html
Sent from the JMeter - User mailing list archive at Nabble.com.
---------------------------------------------------------------------
To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
Re: Multiple Jmeter Instances
Posted by Deepak Shetty <sh...@gmail.com>.
hi
as before you arent able to generate a load of 1000 threads from a single
server. if you post some hardware/OS specs of the machine from which you are
running JMeter (including JVM) we might be able to confirm (you might be
able to get more out of your single instance by following the
recommendations like disabling listeners etc etc)
>they seem to be right. I just wish I understood the reason behind it.
Your client has limited resources. if you ask it to do a lot , then every
thread will have to wait its turn , which means the server sees lower loads.
if like a previous poster mentioned you have a listener writing to your disk
on client then a lot of time goes there so essentially your server sees a
lot less load- if you have any monitoring tool available on your server then
you might check how many threads are active concurrently or you might check
number of active sockets (but this is harder since you will have to check
the state).
You cannot overload your client and have accurate results in terms of time.
regards
deepak
On Mon, Dec 27, 2010 at 11:04 AM, alwaysbmore <al...@yahoo.com> wrote:
>
> Thank you Deepak, your right it's exactly the opposite of what you expect.
> But yes the single instance is providing a lower response times then
> multiples machines.
> I'm starting to lean towards that fact that I need to stick with the
> results
> from multiple machines. Just by observing the site manually during the
> test,
> they seem to be right. I just wish I understood the reason behind it.
> --
> View this message in context:
> http://jmeter.512774.n5.nabble.com/Multiple-Jmeter-Instances-tp3319388p3319626.html
> Sent from the JMeter - User mailing list archive at Nabble.com.
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
>
>
Re: Multiple Jmeter Instances
Posted by alwaysbmore <al...@yahoo.com>.
Thank you Deepak, your right it's exactly the opposite of what you expect.
But yes the single instance is providing a lower response times then
multiples machines.
I'm starting to lean towards that fact that I need to stick with the results
from multiple machines. Just by observing the site manually during the test,
they seem to be right. I just wish I understood the reason behind it.
--
View this message in context: http://jmeter.512774.n5.nabble.com/Multiple-Jmeter-Instances-tp3319388p3319626.html
Sent from the JMeter - User mailing list archive at Nabble.com.
---------------------------------------------------------------------
To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
Re: Multiple Jmeter Instances
Posted by Deepak Shetty <sh...@gmail.com>.
if the problem was that your load wasnt being properly distributed on the
servers due to sticky IP's you would expect the times to be higher on the
single instance than on multiple instances.
1000+ threads will probably not work accurately on a single machine unless
it has good hardware and is well tuned (rarely done for clients) - In all
probability your single machine cant really generate 1000 concurrent
requests - http://wiki.apache.org/jakarta-jmeter/HowManyThreads (follow the
practices listed in lean_mean).
If you have multiple machines , use that - the more you have the more
accurately you can simulate load.
regards
deepak
On Mon, Dec 27, 2010 at 8:34 AM, alwaysbmore <al...@yahoo.com> wrote:
>
> @Adrian Thanks, I will play around with the source IP address. I need to
> do
> some research to figure out how it works.
>
>
> @Felix No, I'm not using constant through put timer, should I?
> --
> View this message in context:
> http://jmeter.512774.n5.nabble.com/Multiple-Jmeter-Instances-tp3319388p3319472.html
> Sent from the JMeter - User mailing list archive at Nabble.com.
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
>
>
Re: Multiple Jmeter Instances
Posted by alwaysbmore <al...@yahoo.com>.
@Adrian Thanks, I will play around with the source IP address. I need to do
some research to figure out how it works.
@Felix No, I'm not using constant through put timer, should I?
--
View this message in context: http://jmeter.512774.n5.nabble.com/Multiple-Jmeter-Instances-tp3319388p3319472.html
Sent from the JMeter - User mailing list archive at Nabble.com.
---------------------------------------------------------------------
To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
Re: Multiple Jmeter Instances
Posted by Adrian Speteanu <as...@gmail.com>.
Hi guys,
In addition:
a. If you suspect that the server behaves differently when receiving
requests from the same IP, you could double check this using the new feature
in version 2.4, ip spoofing (available only for the Http Request Http
Client sampler). Note that some network admins might not be ok with this
practice and it is not recommended unless the test system is isolated from
the rest of the work network.
b. When testing distributed (even with no server-client config), save the
logs (as .csv) to the disk and then load them both, after the test, into a
single & clean listener in order to properly compare with the previous
results (for example, throughput is tricky unless you aggregate all results,
from all jmeter servers instances).
Hope this helps.
On Mon, Dec 27, 2010 at 5:35 PM, Anthony Johnson <an...@gmail.com> wrote:
> Hello,
> 100 threads is not an over-whelming amount of load from my
> experience. The 2 engine results would seem to show that the
> botteneck is somewhere in your test plan.
>
> One common problem is if you use lots of Listeners, I would try
> disabling some and see if your test improves on a single machine. For
> instance, the "View Results Tree" is very useful, but is also very
> heavy. When cutting the test in half and putting on two machines, a
> listener type bottleneck would be lessened with extra CPU and less
> thread contention.
>
> if that doesn't help, you may want to include more details about your
> test, jmeter version, and environment. An outline of the test would
> allow for a lot better insight.
>
> Good luck,
>
> Anthony
>
> On Mon, Dec 27, 2010 at 9:51 AM, alwaysbmore <al...@yahoo.com> wrote:
> >
> > I get two different totally different results. When running multiple
> JMeters
> > instances on multiples machines compared to a single Jmeter instance on a
> > single machine. For example, If I'm looking to run 100 concurrent users.
> If
> > I run all 100 threads from a single machine the results are much lower,
> if I
> > were to run 50 threads on one machine and 50 on another. Both instances
> are
> > run separate from each other, not as a master and server.
> >
> > I suspect, it might have something to do with all my how the server is
> > configured to accept request all coming from a single IP Address.
> >
> > Can anyone one confirm this, and give some feedback on what results
> should I
> > trust?
> > --
> > View this message in context:
> http://jmeter.512774.n5.nabble.com/Multiple-Jmeter-Instances-tp3319388p3319388.html
> > Sent from the JMeter - User mailing list archive at Nabble.com.
> >
> > ---------------------------------------------------------------------
> > 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
>
>
Re: Multiple Jmeter Instances
Posted by Anthony Johnson <an...@gmail.com>.
Hello,
100 threads is not an over-whelming amount of load from my
experience. The 2 engine results would seem to show that the
botteneck is somewhere in your test plan.
One common problem is if you use lots of Listeners, I would try
disabling some and see if your test improves on a single machine. For
instance, the "View Results Tree" is very useful, but is also very
heavy. When cutting the test in half and putting on two machines, a
listener type bottleneck would be lessened with extra CPU and less
thread contention.
if that doesn't help, you may want to include more details about your
test, jmeter version, and environment. An outline of the test would
allow for a lot better insight.
Good luck,
Anthony
On Mon, Dec 27, 2010 at 9:51 AM, alwaysbmore <al...@yahoo.com> wrote:
>
> I get two different totally different results. When running multiple JMeters
> instances on multiples machines compared to a single Jmeter instance on a
> single machine. For example, If I'm looking to run 100 concurrent users. If
> I run all 100 threads from a single machine the results are much lower, if I
> were to run 50 threads on one machine and 50 on another. Both instances are
> run separate from each other, not as a master and server.
>
> I suspect, it might have something to do with all my how the server is
> configured to accept request all coming from a single IP Address.
>
> Can anyone one confirm this, and give some feedback on what results should I
> trust?
> --
> View this message in context: http://jmeter.512774.n5.nabble.com/Multiple-Jmeter-Instances-tp3319388p3319388.html
> Sent from the JMeter - User mailing list archive at Nabble.com.
>
> ---------------------------------------------------------------------
> 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