Detection of network cables getting unplugged

[mushabbar]

Hi,
Inherent problem with TCP/IP sockets is that any network cables unplugged along the route from one side to the other will cause a loss of connection without any notification. To address this issue at the socket level, one recommended approach is to use TCP keepalive mechanism to detect Dead Peers/Process crash/NW cable disconnect. It requires one to enable TCP/IP networking in order to use it. You also need procfs support and sysctl support to be able to configure the kernel parameters at runtime.
In Linux system the following 3 system params needs to be overwritten with user defined values
tcp_keepalive_time (/proc/sys/net/ipv4/tcp_keepalive_time)
tcp_keepalive_intvl /proc/sys/net/ipv4/tcp_keepalive_intvl)
tcp_keepalive_probes /proc/sys/net/ipv4/tcp_keepalive_probes

With the sample prototype setting the above 3 params to 30, 10, 5 was able to detect NW disconnect in ~1.25 min (30s + 10 x 5 = 80s)

ON Solaris platform
To achive the same in Solaris adjusted the foll 2 params
tcp_time_wait_interval
tcp_keepalive_interval

But results are not very satisfying, I’m not able to fine tune system parameters to have the disconnect happen after a specific time. I also noticed that Solaris does not have ‘tcp_keepalive_probes’ so how can one specify the number of re-tries before disconnecting? Basically If I need the socket disconnection to happen in about 80s/100s to what values should I set the Sol NW params?

Thanks in Advance for your help

Regards
Hussain

[Rubberman]

TCP/IP is designed to handle temporary network problems like disconnected cables, etc. That is why the timeout parameters may not seem reasonable to you, but are in the context of how the system was designed/implemented. Consider that TCP/IP is intended to deal with catastrophic events, such as a nuclear attack, and still communicate after such an event... So, the system is functioning as designed.