Suggesting setting a default option to help workaround a issue todo with idle connections on 'home routers'

Binary package hint: procps

A common problem encountered on home routers which behave as NATs is that they tend to drop idle connections (such as those used by ssh, idling at a telnet prompt, older MUD/MUCK/MUCK games and others).

The issue is that home routers tend to assume after a certain amount of time that the connection died and will remove the routing information from memory to free up memory to handle other/more connections. The grace period is usually 900 seconds on home routers that use ipchains internally.

Currently a workaround for this issue exists by tweaking the TCP keep-alive parameter in the client operating systems. The concept is that a keep-alive request is requested before the grace period of the router's routing, the router resets it's garbage collecting timer for that connection and assume it is still active.

By default the Linux kernel uses 7200 seconds (two hours), which is rather far beyond the grace period of most home routers (infact I don't know any that have it set so long) being used as NATs. I would like to suggest that this default value be changed to something much lower like '300', this would prevent a typical user from the frustration of finding he keeps getting disconnected from something just because the connection is idle.

So far I have been unable to find any issues caused by increasing the amount of keep-alives being sent.

Currently the value can be changed on the fly by doing (change 300 to whatever value you like):

sudo bash -c "echo 300 > /proc/sys/net/ipv4/tcp_keepalive_time"

It can also be set as a default option on boot time by editing /etc/sysctl.conf and adding:

net.ipv4.tcp_keepalive_time=300