Hello, I am running a tcp server process on a domU guest machine. Connections are established to the server, payload exchanged and the connection is closed in a non persistent(non keep alive) tcp manner. The client is external to the bare metal machine. Because the connections are ephemeral, 1000s of connections up to an observed peak of ~15500 connections go to TIMED_WAIT state. This is expected behaviour from TCP kernel perspective. However traffic throughput is adversely impacted with the kernel unable to accept connections. Does the VM (domU instance) have access to the full range of ports? I ask because there are 4 domU on the bare metal machines and I was just wondering if 65,000/4 = 16250 is a coincidence being close to the observed peak of ~15500 connections which all require a port of course. Or would you expect that the domU instance has access to the full range of ports? I have tried kernel settings such as tcp_tw_reuse ( just for testing change in behaviour) both on the domU and on the dom0 machines but this did not have any effect. Interestingly, when I place the test client on another of the domU instances, while I can still observe many TIME_WAIT connections, the traffic is not impacted and constant throughput can be achieved. In this case, all traffic is within the bare metal machine. client on dom1 and server on dom2. Would appreciate any thoughts. Thanks.