[Fwd: Re: monitoring with jmeter]

[Jeremy Arnold <je...@bigfoot.com>]

Hello,
    I'm moving this thread to jmeter-dev from jmeter-user so we can 
discuss some of the details.

    Some of the thoughts I mentioned earlier on the JMeter 2.0 thread 
seem to be related to what Peter is talking about doing.  We need to 
have a way to collect statistics other than the things we can learn 
directly from the samplers.  I still haven't thought through the details 
enough, but some general ideas.  Some of this may already be possible.

1) JMeter should provide a way to perform some action at the beginning 
and end of a run, or at periodic intervals.  Perhaps also just after 
some "warm up" period and just before some "cool-down" period.

2) This "action" could be anything, but one common example might be to 
request a web page which would return some statistics about the server 
system, HTTP server, Tomcat, or whatever.  Or perhaps send some data 
over a socket to talk to a simple application on the server which would 
collect data like CPU utilization and paging rates.  Maybe this is 
implemented by executing a specific sampler at the configured times.

3) Results from this arbitrary action would need to be handled on the 
JMeter client.  This probably means having some kind of plugin to JMeter 
that knows what to do with it.  A simple example might be saving the 
contents of the web page to disk.  In some cases the response could be 
ignored altogether.  Or it might parse an XML document and make the 
results available to JMeter to graph alongside normal sampler results.  
This last option will probably require some changes to how we save 
results, but we've talked about changing that anyway.  In my mind, this 
is the hardest step.


Peter: Does this sound similar to what you are trying to do?

Other comments from anybody?

Jeremy

Re: [Fwd: Re: monitoring with jmeter]

[peter lin <jm...@yahoo.com>]

 
yeah that pretty much captures the general scope of it. Given than many servlet and ejb containers support JMX, one thought I had was to alternatively use JMC to poll the server for the data rather than explicitly going to a servlet. 
 
currently the way Tomcat5 handles this internally in the servlet is it uses mbeans. In theory, that could be the mechanism for interfacing with servlet/ejb containers. the catch is non-java servers don't have capability, so using mbeans is J2EE centric.
 
I was talking to remy about different methods to get things like CPU usage and other low level system statistics. Since the VM spec doesn't include lots of other system statistics, like connections, file handles, and CPU utilization, a developer might want to write their own daemon to send that information to a monitor. If we include that mode of monitoring, it might be good to have a monitor which can either poll or simply listen. It could also use the RMI and simply filter out the statistics and feed it to a separate location.
 
 
peter
 


Jeremy Arnold <je...@bigfoot.com> wrote:
Hello,
I'm moving this thread to jmeter-dev from jmeter-user so we can 
discuss some of the details.

Some of the thoughts I mentioned earlier on the JMeter 2.0 thread 
seem to be related to what Peter is talking about doing. We need to 
have a way to collect statistics other than the things we can learn 
directly from the samplers. I still haven't thought through the details 
enough, but some general ideas. Some of this may already be possible.

1) JMeter should provide a way to perform some action at the beginning 
and end of a run, or at periodic intervals. Perhaps also just after 
some "warm up" period and just before some "cool-down" period.

2) This "action" could be anything, but one common example might be to 
request a web page which would return some statistics about the server 
system, HTTP server, Tomcat, or whatever. Or perhaps send some data 
over a socket to talk to a simple application on the server which would 
collect data like CPU utilization and paging rates. Maybe this is 
implemented by executing a specific sampler at the configured times.

3) Results from this arbitrary action would need to be handled on the 
JMeter client. This probably means having some kind of plugin to JMeter 
that knows what to do with it. A simple example might be saving the 
contents of the web page to disk. In some cases the response could be 
ignored altogether. Or it might parse an XML document and make the 
results available to JMeter to graph alongside normal sampler results. 
This last option will probably require some changes to how we save 
results, but we've talked about changing that anyway. In my mind, this 
is the hardest step.


Peter: Does this sound similar to what you are trying to do?

Other comments from anybody?

Jeremy



> ATTACHMENT part 2 message/rfc822 name=Re: monitoring with jmeter
Date: Thu, 28 Aug 2003 05:53:43 -0700 (PDT)
From: peter lin 
Subject: Re: monitoring with jmeter
To: JMeter Users List 


I'm actually working on a jmeter monitor for Tomcat. So far I've been focusing on Tomcat, since well I use Tomcat. Mike and I have been discussing this for a couple weeks now and I'm also talking to remy about providing support in Tomcat to facilitate this.

my plans so far is this.

1. update the new TC5 status servlet to output XML
2. submit the patch once remy likes it. since I need his help to convince the other committers to Ok the patch
3. write a JMeter monitor that is configurable

The type of monitor I had in mind is more general. What I want to do is be able to build tomcat + webapps nightly and deploy it. Then an Ant process would kick off to load and regression test the changes. The monitor would ping the server on a set interval to see how it performs. The information provided by the status servlet in TC5 include threads, vm and so on. The design I had in mind was to have a generic parser interface, which the monitor uses. that way it maintains the same level of flexibility and extensibility as other Jmeter components. The regression/stress test plan would have one monitor that records the results. Time permitting after all that is done, I want to write a GUI that allows me to view the performance of the server for the duration of the test.

Does that sound similar to what you were thinking of?

peter


erik.itter@bbraun.com wrote:
Hi,

I thought about monitoring our web farm using JMeter. I would write a 
console to view all unsucessfull polls on one spot and extend the 
listeners. Of course that console would need to do alerting as well but 
thats somewhat out of scope... (and quite easy using java)

Has anyone done such a thing so far?
Is the whole idea nonsense?
Is JMeter stable enough to run infinite?
Does JMeter work in an environment where I need to test about 1000 targets 
each minute?

any comments welcome

---------------------------------
Do you Yahoo!?
Yahoo! SiteBuilder - Free, easy-to-use web site design software---------------------------------------------------------------------
To unsubscribe, e-mail: jmeter-dev-unsubscribe@jakarta.apache.org
For additional commands, e-mail: jmeter-dev-help@jakarta.apache.org

---------------------------------
Do you Yahoo!?
Yahoo! SiteBuilder - Free, easy-to-use web site design software