TCPIP socket did not drop

[Dushyant_2]

I have server running OpenVMS 7.3-1 running Decnet V and TCPIP version Open VMS Alpha version V5.3 running. My problem was if one TCPIP device disconnect from the network then this server is not dropping that socket for around 8.3 minute to 9 minute. I tried so many thing after that I got suggestion from HP for enable Keep Alive timer, now it's around 90 Sec. But it's still bugging me "why and what" is magic timer for 8.3 minute to 9 minute. And why that socket did not drop? This is happing if I remove that server from network also that means; If I have five TCPIP connection to this server and I remove network cable from this server all five socket will not drop for around 9 minute.

Any body has experiences this kind of problem?

[Tim Nelson]

I have experienced and seen this item on other platforms.

If a network connection drops, compared to exiting, the OS will wait for the duration of the keep alive timer before sending the keep alive query or closing the socket. On some OS's this can be up to 2 hours.

The keep alives exist for idle connections, once the line has been idle for the keep alive interval the OS will send a query. If someone answers from the other end the timer starts over. If no one answers the steps to close the port are begun.

Luck

[labadie_1]

Read the funny manual about Tcp keepalive at (for Tcpip 5.3)

http://h71000.www7.hp.com/doc/73final/6631/6631pro_003.html#keepalive

And to see your parameters, do
$ @sys$startup:tcpip$define_commands
$ sysconfig -q inet

[Ian Miller.]

there was a big change (to make the relevent more standard apparently) in recent versions which made this time much longer. Note to can't these things permentently but you have to put the netconfig commands in the startup or there may be another file.

[Willem Grooters]

8.3 minutes = about 500 seconds.

ANY connection can break, intentionally or accidently. In the early days of networking, it would be a rather normal practice that physical connections were lost for a longer or shorter period. It the system would react immediately, it would need to crash the programs using the connection, and these would need to be restarted as soon as the connections would be up again - maybe tenths of seconds, or some seconds later, and having the chance to crash them dureing starup because the connections were lost again.
By allowing outage for some time, this would be avoided. Any period is arbitrary. It can be chnaged (check the TCPIP manuals) but it would be changed for ALL sockets. This might be a problem...

Willem