It has taken us 6 months to resolve those networks troubles. The users were usually experiencing TSE disconnection/freezes but not only. All the networks services were sometimes running at half their normal speed. That was quite complicated to find the sources of our problem because the network had changed a lot at this time (new server, wi-fi network, new persons in charge…).
After a while, I decided to look carefully the packet stream thanks to a port mirroring on different servers. All of our servers were running windows server 2003.
I have found that kind of comportment:
Captured with Wireshark
Then, my colleague Pascal Medi wrote a short program to calculate the proportion of bad packets (I will publish it on the 9th of december).
In our case, they were too many WiFi users who were connected on a low signal quality connection. That’s why they needed tcp restransmissions and tcp window updates.
The conclusion is that, if you are experiencing network performance problems with a windows server 2003 infrastructure, don’t forget to verify your low layer…