We have a two node TMG implementation running SP1. The first node (10.0.1.9) runs like a champ. Node two (10.0.1.10) runs OK but then performance nose-dives eventually leading to a completely non-functional server. We have found through netstat that thousands and thousands of ports on node two are tied up in a TIME_WAIT to the cluster IP address (10.0.1.8) and eventually there are no ports available to serve new connections. We have tried increasing the ports and decreasing the timeout but neither seems to be working. Any ideas? Any questions I can answer to move this along?