[JDEV] Heartbeat patch for dialup and laptop users and faulty presence info

[Bernd Eckenfels]

On Tue, May 28, 2002 at 01:49:45PM -0600, David Waite wrote:
> Are you thinking of having an explicit heartbeat required for clients to 
> send over the network socket? I'm not sure if I agree with circumventing 
> the TCP timeout with application logic or not - it seems like just 
> giving a delivery error once the socket closes would be a better 
> long-term solution.

The problem is, that the socket is not closed on a hard physical disconnect
(in some conditions). If no TCP RST Packet (because of a rebooted machione
or a new client with the old ip) or ICMP Error (i.e. no route to destination
from dialup server) is generated, TCP will happyly retransmit packets for
minutes. It is even worse on the client side, if the client does not send
anything to the server, expecting the server to send news. The client socket
is alive, the server socket is reset. since the server does not tell the
client so, your client wont noticve that he is dead.

A TCP ping is one thing, to detect reset sockets which are idle. On IRC for
example in addition to the ping, there is also a pong message required
within a (configurable time frame). The server will simply disconnect the
connection of no repsonse is received. this requires a echo protocol
function but is IMHO quite important.

Clients could even be so smart and ping the server and measure roundtrip or
connect resets sthemself.

Greetings
Bernd