jmeter_distributed_testing_step_by_step
jmeter_distributed_testing_step_by_step
This short tutorial explains how to use multiple systems to perform stress testing. Before we start,
there are a couple of things to check.
Once you've made sure the systems are ready, it's time to setup remote testing. The tutorial
assumes you already have JMeter installed on all the systems. The way JMeter works is 1 master
controller initiates the test on multiple slave systems.
Diagram 1
Terminology
Before we dive into the step-by-step instructions, it's a good idea to define the terms and make sure
the definition is clear.
Master – the system running Jmeter GUI, which controls the test
Slave – the system running jmeter-server, which takes commands from the GUI and send requests to
the target system(s)
Target – the webserver we plan to stress test
1/4
Step-by-Step
2/4
Starting the Test
At this point, you are ready to start load testing. If you want to double check the slave systems are
working, open jmeter.log in notepad. You should see the following in the log.
If you do not see this message, it means jmeter-server did not start correctly. For tips on debugging
the issue, go to the tips section. There are two ways to initiate the test: a single system and all
systems.
3/4
Limitations
There are some basic limitations for distributed testing. Here's the list of the known items in no
specific order.
1. RMI cannot communicate across subnets without a proxy; therefore neither can jmeter without a
proxy.
2. Since JMeter sends all the test results to the controlling console, it is easy to saturate the
network IO. It is a good idea to use the simple data writer to save the results and view the file
later with one of the graph listeners.
3. Unless the server is a large multi processor system, in most cases 1-2 clients is sufficient to
overwhelm the server.
4. A single JMeter client running on a 2-3Ghz CPU (recent cpu) can handle 300-600 threads
depending on the type of test. (The exception is the webservices). XML processing is CPU
intensive and will rapidly consume all the CPU cycles. As a general rule, the performance of XML
centric applications will perform 4-10 slower than applications using binary protocols.
Additional resources
http://wiki.apache.org/jmeter/JMeterFAQ#How_to_do_remote_testing_the_.27proper_way.27.3F
http://jmeter.apache.org/usermanual/remote-test.html
Tips
Windows firewall
Linux
On Suse linux, ipchains is turned on by default. For instructions, please refer to the “remote testing”
in the user manual.
On RedHat (or derivatives), iptables (netfilter) is turned on by default. Execute “service iptables
stop” to stop the Linux netfilter firewall.
4/4