Discussion:
stress testing with ramp-up and time periods (LDAP/SLAMD)
Quanah Gibson-Mount
2018-10-10 23:31:47 UTC
Permalink
Hi,

In the past I've used slamd to stress test various LDAP servers to
determine what the max throughput is that they could handle for various
operations or sets of operations. However, in reading the jmeter docs, I'm
not clear how one would replicate this type of testing.

Generally, in SLAMD, I could configure things to behave in the following
way:

1) Set up your distributed clients (as can be done with jmeter)

2) Configure the overall system so that each client starts out with
initially 1 thread executing your task, with W amount of warmup time, and
then a duration of X, with a cooldown of Y seconds afterward. Then
increment the thread count on each client, and do the same thing. You do
this until there is no performance increase for at least Z iterations.
Once that maximum is determined, re-run the best iteration for a configured
amount of time.

For example, here's the generated report for a stress test I ran several
years ago:

<https://mishikal.files.wordpress.com/2013/05/mdb_slamd_data_report.pdf>

In this case, I had 16 distributed clients. They would run each iteration
for 1200 seconds, with a 30 second delay between iterations. They would
start with 1 thread each, and no specified maximum (since this is
controlled by the improvement counter). There was a warmup of 60 seconds
before statistics would be gathered and the iteration timer was started and
a cooldown of 60 seconds at the end of the iteration. Statistics are
collected every 10 seconds.


Does anyone have pointers and/or documentation that would allow me to set
up a similar sort of stress test with jmeter as slamd was abandoned several
years ago?

Thanks,
Quanah

--

Quanah Gibson-Mount
Product Architect
Symas Corporation
Packaged, certified, and supported LDAP solutions powered by OpenLDAP:
<http://www.symas.com>


---------------------------------------------------------------------
To unsubscribe, e-mail: user-***@jmeter.apache.org
For additional commands, e-mail: user-***@jmeter.apache.org
Felix Schumacher
2018-10-13 13:55:48 UTC
Permalink
Post by Quanah Gibson-Mount
Hi,
In the past I've used slamd to stress test various LDAP servers to
determine what the max throughput is that they could handle for
various operations or sets of operations.  However, in reading the
jmeter docs, I'm not clear how one would replicate this type of testing.
Generally, in SLAMD, I could configure things to behave in the
1) Set up your distributed clients (as can be done with jmeter)
2) Configure the overall system so that each client starts out with
initially 1 thread executing your task, with W amount of warmup time,
and then a duration of X, with a cooldown of Y seconds afterward. Then
increment the thread count on each client, and do the same thing.  You
do this until there is no performance increase for at least Z
iterations. Once that maximum is determined, re-run the best iteration
for a configured amount of time.
For example, here's the generated report for a stress test I ran
<https://mishikal.files.wordpress.com/2013/05/mdb_slamd_data_report.pdf>
In this case, I had 16 distributed clients.  They would run each
iteration for 1200 seconds, with a 30 second delay between
iterations.  They would start with 1 thread each, and no specified
maximum (since this is controlled by the improvement counter). There
was a warmup of 60 seconds before statistics would be gathered and the
iteration timer was started and a cooldown of 60 seconds at the end of
the iteration.  Statistics are collected every 10 seconds.
Does anyone have pointers and/or documentation that would allow me to
set up a similar sort of stress test with jmeter as slamd was
abandoned several years ago?
I have to admit, that I didn't completely understood how you wanted to
accomplish with your setup.

With JMeter you can use ramp-up period out of the box, that will start
one thread after another over the specified duration. You can limit the
overall duration of the test by using the scheduler of the default
thread group.

If you need more you can look for JMeter plugins.
Have you had a look at the JMeter plugin
https://jmeter-plugins.org/wiki/ConcurrencyThreadGroup/?

For the reports have a look at the report generation
http://jmeter.apache.org/usermanual/generating-dashboard.html

Regards,
 Felix
Post by Quanah Gibson-Mount
Thanks,
Quanah
--
Quanah Gibson-Mount
Product Architect
Symas Corporation
<http://www.symas.com>
---------------------------------------------------------------------
---------------------------------------------------------------------
To unsubscribe, e-mail: user-***@jmeter.apache.org
For additional commands, e-mail: user-***@jmeter.apache.org

Loading...