Setting nb threads / ramp-up period / loop count, for a long simulation

[Mathieu Seillier <st...@gmail.com>]

Hello,

I would like some clarification about the setup for my test plan.
First of all, I explain quickly the behavior of my system :
- Each user of the system performs operations on his station (heavy client)
- One operation for one user includes data entry on the station and 5
requests to some web services of my system
- so in my Jmeter script, my test plan contains 5 SOAP requests, and before
each requests I have added 60sec timers, to simulate the work of a user on
his station.
- A complete operation for a user then takes about 5 minutes (1 min before
each web service request).
My system must support (max load) :
- Up to 40,000 transactions / day
- 1 day = 8 hours, so up to 5,000 operations / hour (no load peak, it is
estimated that the operations are almost evenly distributed over the day)
- About 500 simultaneous users (the number of users remains almost constant
all day)
- so, it gives 10 transactions per user per hour.

For the number of threads, it's easy, it must equal 500 because it
corresponds to the users of my system.
On the other side, for the Ramp-up period, as I understand it's the time to
reach the max number of users.
In my case, the nb of users is almost the same all day, does that mean I
have to put "1" for the ramp-up period ?
Also how can I simulate a test day during 8 hours? Is the scheduler (with
start time and end time) useful for that ?
How do I set my script to set up 10 operations per user per hour ?

Thank you for your help
stieuma

Re: Setting nb threads / ramp-up period / loop count, for a long simulation

[sebb <se...@gmail.com>]

On 2 December 2011 17:26, Tanakiat Srisaranayakul <ta...@gmail.com> wrote:
> Please see my answer inline.
>
> In my case, the nb of users is almost the same all day, does that mean I
>> have to put "1" for the ramp-up period ?
>>
> If the machine running jmeter is not good enough, 500 thread cannot be
> started in 1 second. Are you ok if it's not 500 concurrent thread at 1st
> sec?

Even with a longer ramp-up, 500 threads may be too much for one JMeter
or even a single host; might need 2 or 3.

You can check the absolute limit for JMeter by replacing your samples
with a Java sampler, or perhaps an HTTP sampler reading a file.
If you have a fast local server, you could also test against that,
which will help show any limitations on network resource usage.
Use a similar amount of data per request if possible.

However, by reducing the timeouts, one can reduce the number of
threads but still maintain the same load on the server.
[JMeter can generate requests much faster than a human]

The important measure here is the request rate as seen by the server under test.
Whether it is generated by lots of single users or fewer faster
threads probably won't affect the server response time as much as the
total load.
You can certainly get useful information from such a test.

If you want to test with more users, you will most likely need to use
multiple JMeters and hosts.

> Also how can I simulate a test day during 8 hours? Is the scheduler (with
>> start time and end time) useful for that ?
>>
> If you'd like to test for 8 hour and 1 time, it's ok to use jmeter
> scheduler. If you'd like to run some kind of longevity test for few days,
> you may need to consider OS scheduler to execute the script daily.
>
> How do I set my script to set up 10 operations per user per hour ?
>>
> Throughput Controller should fit your requirement
> http://jmeter.apache.org/usermanual/component_reference.html#Throughput_Controller

---------------------------------------------------------------------
To unsubscribe, e-mail: user-unsubscribe@jmeter.apache.org
For additional commands, e-mail: user-help@jmeter.apache.org

Re: Setting nb threads / ramp-up period / loop count, for a long simulation

[Tanakiat Srisaranayakul <ta...@gmail.com>]

Please see my answer inline.

In my case, the nb of users is almost the same all day, does that mean I
> have to put "1" for the ramp-up period ?
>
If the machine running jmeter is not good enough, 500 thread cannot be
started in 1 second. Are you ok if it's not 500 concurrent thread at 1st
sec?

Also how can I simulate a test day during 8 hours? Is the scheduler (with
> start time and end time) useful for that ?
>
If you'd like to test for 8 hour and 1 time, it's ok to use jmeter
scheduler. If you'd like to run some kind of longevity test for few days,
you may need to consider OS scheduler to execute the script daily.

How do I set my script to set up 10 operations per user per hour ?
>
Throughput Controller should fit your requirement
http://jmeter.apache.org/usermanual/component_reference.html#Throughput_Controller

Re: Setting nb threads / ramp-up period / loop count, for a long simulation

[ZK <st...@gmail.com>]

>For the number of threads, it's easy, it must equal 500 because it 
>corresponds to the users of my system. 

Why not consider 50 threads doing 1 request every 6 seconds?



ZK

--
View this message in context: http://jmeter.512774.n5.nabble.com/Setting-nb-threads-ramp-up-period-loop-count-for-a-long-simulation-tp5042325p5043143.html
Sent from the JMeter - User mailing list archive at Nabble.com.

---------------------------------------------------------------------
To unsubscribe, e-mail: user-unsubscribe@jmeter.apache.org
For additional commands, e-mail: user-help@jmeter.apache.org

Re: Setting nb threads / ramp-up period / loop count, for a long simulation

[sergio <se...@bosoconsulting.it>]

Hi, Mathieu

it is never a good idea to put "1 second" in the ramp-up.
Even if your Jmeter server would be able to start 500 thread in a single second (I think you can do it with large server),
you will end up with 500 threads generating their requests almost at the same time.
This is a very, very worst case which probably will mess up your target server.

So, you can distribute the requests by setting a ramp-up time similar to your cycle time.
As a second choice, you can add a linear random timer at the beginning of each thread.

best regards
Sergio

Il 02/12/2011 16:59, Mathieu Seillier ha scritto:
> Hello,
>
> I would like some clarification about the setup for my test plan.
> First of all, I explain quickly the behavior of my system :
> - Each user of the system performs operations on his station (heavy client)
> - One operation for one user includes data entry on the station and 5
> requests to some web services of my system
> - so in my Jmeter script, my test plan contains 5 SOAP requests, and before
> each requests I have added 60sec timers, to simulate the work of a user on
> his station.
> - A complete operation for a user then takes about 5 minutes (1 min before
> each web service request).
> My system must support (max load) :
> - Up to 40,000 transactions / day
> - 1 day = 8 hours, so up to 5,000 operations / hour (no load peak, it is
> estimated that the operations are almost evenly distributed over the day)
> - About 500 simultaneous users (the number of users remains almost constant
> all day)
> - so, it gives 10 transactions per user per hour.
>
> For the number of threads, it's easy, it must equal 500 because it
> corresponds to the users of my system.
> On the other side, for the Ramp-up period, as I understand it's the time to
> reach the max number of users.
> In my case, the nb of users is almost the same all day, does that mean I
> have to put "1" for the ramp-up period ?
> Also how can I simulate a test day during 8 hours? Is the scheduler (with
> start time and end time) useful for that ?
> How do I set my script to set up 10 operations per user per hour ?
>
> Thank you for your help
> stieuma
>

L’interessato può, inoltre, esercitare tutti i diritti di accesso sui propri dati previsti dall’art.7 del d.lgs.vo n. 196/2003, tra 
i quali i diritti di rettifica, aggiornamento e cancellazione, inviando un messaggio all’indirizzo:sergio@bosoconsulting.it 
<ma...@yahoo.it>
In caso di erronea ricezione da parte di persona diversa, siete pregati di eliminare il messaggio e i suoi allegati in modo 
definitivo dai vostri archivi e di volercelo comunicare immediatamente restituendoci il messaggio via e-mail al seguente 
indirizzosergio@bosoconsulting.it <ma...@yahoo.it>





---------------------------------------------------------------------
To unsubscribe, e-mail: user-unsubscribe@jmeter.apache.org
For additional commands, e-mail: user-help@jmeter.apache.org