Can OpenSTA test the web stress with ramp up users...
I am exploring whether OpenSTA can be the best free tool to test stress/performance on a web form.
Our scenario is 50 concurrent usrs log on a web form and load the huge data(actaully it's a activeX embedded), then ramp up 5 users every 10mins say, the target is to make sure the query time within a reasonable range.
Can OpenSTA help us finish such web stress? Thanks in advance for any ideas.
Re: Can OpenSTA test the web stress with ramp up users...
It really depends on your application under test and the result you want to achieve.
Off the top of my head I can't remember if ActiveX can do this, but, if your ActiveX is being used to open a separate connection to the server (e.g. for non-http traffic), then you are going to have a problem finding any free tool that will handle this. For that matter, most commercial tools won't be able to handle it. LoadRunner can but its prohibitively expensive (even for zillionaires).
However, if all traffic between the browser and server is http, then OpenSTA is a good tool to perform normal performance testing. It can easilly handle 50 concurrent users (I think there is an open bug that says there is a problem doing >1600 users on some platforms, but, other than that the max is determined by the network bandwidth on your load generator). It can also handle the ramping up (although ramping up must be determined prior to test execution - there is no way to add/remove VUsers dynamically during test execution).
The impact to the tester is that, although the target server is under the correct amount of load and performing as it would be if the requests were coming from thousands of real users, the response times being measured are calculated from the beginning of the request being sent out to the end of the response being received. This may not be what the user would experience. If you have a page heavy in JS, or a large ActiveX control, the client machine may take some time to render it. So, the user experience would be response time plus render time.
The only way to properly test this is, if you have some idea of what the configuration of the client machine will be, you can run a 'normal' performance test and, at the same time, either using a functional testing tool, or manually, time end-to-end user experience on a separate client machine.
Hope this helps,