network layer hangs (TX stops) when it's slammed w/ a stress test
I'm running on an XP host with the max socket# set for 32K instead of the default 5000. I configured a VirtualBox VM running Win2003 and make sure to turn off SynAttackProtect so Windows 2003's network layer isn't shutting off.
The MSSQL server runs on XP. The Win2003 VM runs IIS and my web app but it uses the MSSQL server on the XP VirtualBox host. The VM is set to use the Host Interface for networking so it has an actual IP address on the network.
The OpenSTA stress tester runs on the XP machine. I set up 1000 virtual users to run the captured stream back to the web app and it works ok up to about 500 users. After that, the response time first spikes hard up to 10sec, then the network layer falls over (can't ping from XP machine to Win2003 VM or vice versa)...restarting the VM fixes this network layer hang.
If I bring up the Session Information for the VirtualBox VM, I can see the Receive Byte count go up, but Transmit doesn't change.
FWIW, VMWare scales up to 800ish users before OpenSTA craps out and shuts down w/o saving any results (VMWare's networking stays up). VirtualBox scales up faster, before the networking layer hangs :-)
I've checked for tickets with "network hang" w/o luck, so hopefully this isn't a duplicate. This is the first issue I've found w/ the 2.1 release.