| Author |
Message
|
| hasselb |
 Posted: Thu May 24, 2007 2:57 am Post subject: Simulate a workload with a flow java compute node sleep() ? |
 |
|
Newbie
Joined: 26 Apr 2007
Posts: 6
|
We want to build a flow that simulates a workload. The flow should be waiting for a certain amount of time. Our first idea was , that we use a java compute node with Thread.sleep(interval). Now , we know , that there are problems with this approach. especially we use this test for performance testing with 500 user.
Can anyone give us some hints or a strategy , how we can create a stable solution for that?
Many thanks. |
|
| Back to top |
|
 |
| marcin.kasinski |
| Posted: Thu May 24, 2007 3:02 am Post subject: Re: Simulate a workload with a flow java compute node sleep |
|
|
Sentinel
Joined: 21 Dec 2004
Posts: 850
Location: Poland / Warsaw
|
Have you considered application which sends lots of test messages to Broker flow input queue.
I think it is common way to prepare stress tests.
_________________
Marcin |
|
| Back to top |
|
|
| hasselb |
| Posted: Thu May 24, 2007 3:07 am Post subject: |
|
|
Newbie
Joined: 26 Apr 2007
Posts: 6
|
Yes , this is partially a stress test. We want to perform a performance test for an aplication that uses our message broker. A lot of user put searches over the internet in our system. We have to simulate this. The Message broker flow should simulate some work load within the broker. That is the reason for a "random" delay in the flow stub. For the random delay we wanted to use a configuration file, also o reason for a java compute node.
How can we do it better ?? |
|
| Back to top |
|
|
| marcin.kasinski |
| Posted: Thu May 24, 2007 3:16 am Post subject: |
|
|
Sentinel
Joined: 21 Dec 2004
Posts: 850
Location: Poland / Warsaw
|
| hasselb wrote: |
Yes , this is partially a stress test. We want to perform a performance test for an aplication that uses our message broker. A lot of user put searches over the internet in our system. We have to simulate this. The Message broker flow should simulate some work load within the broker. That is the reason for a "random" delay in the flow stub. For the random delay we wanted to use a configuration file, also o reason for a java compute node.
How can we do it better ?? |
But isn't it better to move this simulation outside broker ?
I think it is better to use test application for that and leave your flow unchanged.
You can send test messages from this application with "random" delay to simulate user behaviour.
I hope I understand your requirements.
_________________
Marcin |
|
| Back to top |
|
|
| marcin.kasinski |
| Posted: Thu May 24, 2007 3:23 am Post subject: |
|
|
Sentinel
Joined: 21 Dec 2004
Posts: 850
Location: Poland / Warsaw
|
Another thing is what causes your application to send message to broker queue.
If it is link maybe it is better to use application which generate URL request with random delay.
What do you want to test , only flow or business funcionality ?
_________________
Marcin |
|
| Back to top |
|
|
| hasselb |
| Posted: Thu May 24, 2007 3:37 am Post subject: |
|
|
Newbie
Joined: 26 Apr 2007
Posts: 6
|
this is a big test szenario. normaly the broker sens the messages to legacy systems. Here we replace the legacy systems with our flow as a stub and simulate some workload from the legacy systems. so we do performace tests from http page to the stubs. we do this with one hour and with 500 simulated user and also the requests are not identical.
we want to know, if our aplication is performant without the legacy systems. |
|
| Back to top |
|
|
| marcin.kasinski |
| Posted: Thu May 24, 2007 4:31 am Post subject: |
|
|
Sentinel
Joined: 21 Dec 2004
Posts: 850
Location: Poland / Warsaw
|
| hasselb wrote: |
this is a big test szenario. normaly the broker sens the messages to legacy systems. Here we replace the legacy systems with our flow as a stub and simulate some workload from the legacy systems. so we do performace tests from http page to the stubs. we do this with one hour and with 500 simulated user and also the requests are not identical.
we want to know, if our aplication is performant without the legacy systems. |
I don't know, but I can not see benefits from using "flow" to simulate legacy application.
Test environment should be similar to PRD. I've never seen DEV and PRD identical, but they shooul be similar. Look at you modifications.
You remove legacy application and modify broker by addind additional flow.
It influences your tests.
In my opinion it is better to prepare stub legacy application rather than flow.
_________________
Marcin |
|
| Back to top |
|
|
|
|
