Hello Mike,
thank you for your reply. As a rule of thump, 1 cpu for 25 users seems to be a good starting point.
My concerns regarding performance settle largely on the RDP hosts - which still leaves me with the problem of creating realistic load on these hosts.
Unfortunatly, the Sikuli project was a bit difficult to find: http://sikuli.org hangs on loading, http://sikulix.org does not exist any more, but alas, I found http://sikulix.com!
As the docs state, it focuses on seeing the browser contents as an image, with OpenCV tools to analyze it.
For the given scenario, that is exactly what is needed (as opposed to the current mainstream tools, which operate on the DOM).
Now I have a tool, but still need a "workbench" - a little fleet of load generating machines where the yet to be taylored Sikuli script would run.
My hopes where that there is some service out there which I could use (e.g. flood.io, loadninja.com, smartmeter).
I tried LoadNinja.com, which allows recording a session and playing it back on simulated browsers.
But the playback seems to be a "blind flight" - it seems to just play back the mouse clicks after the recorded delays, regardless if the UI element that was shown on the original recording it already at the specified place.
So if the response times slow down, the clicks go astray and the playback will do something not reproducable.
LoadNinja allows entering checks along the recording - but these are tayloured for DOM interactions.
Furthermore the recording timeline does not show the screen contents at the recording step - so even if there are screenshot and image recognition tools available in the validation steps, it will be quite tedious. Is Sikuli more comfy in that regard?
Nothing is ever easy.
Best regards,
Bernd
