[prev in list] [next in list] [prev in thread] [next in thread] 

List:       jmeter-user
Subject:    Re: Performance of Jmeter itself
From:       Adrian Speteanu <asp.adieu () gmail ! com>
Date:       2015-07-16 12:22:30
Message-ID: CANEEAcz=YbKiurrcjFqPtvT3Q+cv59NbbGBb5gcVw8cY27A1LA () mail ! gmail ! com
[Download RAW message or body]


How JMeter sends requests is explained in the online documentation and you
have a correct understanding of that. Without any timers and controllers
impacting throughput and wait periods, each thread will attempt another
request as soon as it receives the response from it's previous request!

This becomes a problem only in the case of resource starvation, a large
thread pool for example will add to the operations that a CPU has to
perform and threads that don't receive processor time will obviously have
to wait until they do. It's also a false positive, sometimes, in the case
of applications with large response times per request. For example, we've
ran into a known issue of the Apache server at one point and every
hundredth request or so was experiencing large response times (from 0.5s of
the 90% percentile to 40+ seconds). Obviously, we first noticed that "not
all threads were active" - a false assumption given by the fact that we
were performing less requests per second than we were expecting. With a low
number of active threads in a test, this becomes an issue that keeps a
thread apparently "inactive" (in terms of requests sent, as the thread is
very much alive and functioning properly) for as long as it takes for the
server to respond. But requests are generated sequentially and that thread
needs to wait for the response to arrive completely before proceeding. This
is covered in the response times though, so you should be able to see when
something unusual is going on.

So try experimenting without that plugin and see if you have issues that
are related to the plugin, behaviour should be similar. What I try to do
when I'm experiencing bad behaviour from the server, I increase the thread
pool available and use the Constant Throughput Timer to generate a constant
rate of requests per minute.

Cheers,
Adrian S

On Tue, Jul 14, 2015 at 5:28 PM, George <bamboocha324@yahoo.de> wrote:

> Hi,
> well yea for sure JMeter need "time" to create the socket the connection
> and sent the data - clearly.I do also accept the "think times" of
> JMeter.What i mean is if we can say that JMeter is basically sending the
> request out as fast as it is possible for JMeter - right?So there is no
> "speed up" value.If i want to stress my server i just need to multiple the
> load by putting more threads or seting up more jmeter instances and
> multiple the load with every instance.This is also clear for me.
>
> Of course i noticed that if i set the "connection timeout" values to e.g.
> 1ms than the requests goes out very fast but they are all red (fail)
> because telling jmeter to close the connection after 1ms is very stupid :)
> BrGeorge
>
>
>
>      Adrian Speteanu <asp.adieu@gmail.com> schrieb am 16:14 Dienstag,
> 14.Juli 2015:
>
>
>  Hi,
>
> I had a test script of a very small server requests (pinging the webapp
> basically). JMeter recorded 5ms average response times, monitoring on the
> server side showed 2ms average response times. The difference is
> justifiable due to the load-balancer in between the test client machine and
> the targeted server + jmeter's own "think times". Anything at the level of
> the millisecond is easily considerable acceptable.  What do you mean by
> speeding up?
>
> Cheers,
> Adrian
>
> On Tue, Jul 14, 2015 at 3:55 PM, George <bamboocha324@yahoo.de> wrote:
>
> > Hello,
> > i have a question about JMeter's performance.
> >
> > I have a simple test plan with one thread looping 1000 times
> > I have one SMTP sampler targeting my smtp server on port 25 and sending a
> > file of couple Kb.I have no connection timeout or connection close
> defined.
> > Technically everything works perfect.
> > I know i could put some "timer" or "sleep" to slow down sending my smtp
> > request e.g. i put a "sleep 10" and jmeter waits 10 sec between every
> smtp
> > sampler, but what about to "speed up" the sending process?I mean how
> "fast"
> > do JMeter sent out the smtp request? Is it related to my environement?
> > Can i "speed up" jmeter somehow?
> > I have the same question for http as well.
> > BrGeorge
> >
> >
> >
>
>
>
>


[prev in list] [next in list] [prev in thread] [next in thread]