I created a script (the exact script is below) which is supposed to
test how many connections the server can handle at one time. The
problem is, after a while, performer's connection starts timing out,
even though it's still possible to connect to the server.
I'm running the script on an NT 4.0 machine with service pack 6, 512Mb of RAM, and using Silk Performer 3.5. The server is a linux box.
I looked at the Segue Performer Whitepapers and added the registry key
that is supposed to allow NT to have more connections (the papers said
to set the number to 65534). So that shouldn't be the problem. The
exact error it gave (which always happened at 1010 connections) is
Winsock 10060 - Connection timed out, fake.unreal.com 9999
I think, since it aways happened at 1010, that there must be some limit
to the number of connections either performer or NT can handle. My
question is, I need to know if the problem is performer, NT, or the
Normal 10/100 NICs can handle, even on super mega-server machines, 1000 virtual user connections, then you start timing out. Interval doesn't seem to matter. You're running 1000 users but each with 10 connections, so the 1001st kicks in the limitation. Don't know if that is some international standards-induced phenomenon or not.
The other thing to look for: how many software port addresses are available on the machine? There's a finite number set in most server boxes and workstation boxes, even for Linux.
I think the former might be the case, though.
Sr. QA Group Lead
Intellocity USA, Inc.
I have similar problems when testing a website running on a Windows 2000/IIS5 server.
In my case, I don't get the same errors when testing against a NT4/IIS4 server.
Do you get Winsock errors right away? Or after a period of time? The tests I've been running start with a small number of users (50) and ramp up to 200. I normally see Winsock errors pop up around 120 users.