When an SNA client computer has a TCP/IP socket connection to an SNA Server, the server periodically sends a Keep Alive message to the client to speed up the detection of a lost client LAN session. SNA Server TCP/IP Keep Alive messages do not complete because the client's TCP receive window size dropped to zero. The SNA Server keeps on queuing new Keep Alive messages (even though earlier keep alive messages haven't completed) until the SNA Server runs out of system PTEs.
Resolution:
Microsoft has modified the files, SNADMOD.DLL and SNAIP.DLL, so that SNA Server does not queue a new Keep Alive until the previous one has completed. Thus, the connection is aborted after a configurable timeout. The timeout is specified in the following registry parameter:
The KeepAliveLimit specifies how many WatchDogTimeouts SNA Server waits before aborting the connection. The default watchdogtimeout value is 60 seconds, though this is also configurable under:
If a second keepalive is sent to a client box, SNA Server now logs Event 700 to indicate this (logged only when detailed problem analysis logging is enabled). If the limit is exhausted, then SNA Server logs a 701 event at a high severity to indicate this. The client's IP address is logged in these new events.
The registry value, HardClse, controls how the connection is aborted. If the value is "yes", the connection is reset immediately and no effort is made to close the connection gracefully. The default value is "no". HardClse is found in the following subtree:
