Subject: Re: TCP connections
To: None <current-users@NetBSD.ORG>
From: Scott L. Burson <gyro@zeta-soft.com>
List: current-users
Date: 02/12/1996 11:49:11
   Date: Mon, 12 Feb 1996 08:42:55 +0100
   From: Niklas Hallqvist <niklas@appli.se>

   The other side of the coin is that the server side of the rlogin
   protocol issues keepalive messages, thus forcing the above retransmits
   to happen if the line is down.  This is easy to turn off, just add the
   "-n" option to the rlogind invocation in inetd.conf.  This should make
   the problem less severe even though it's still there.

Ah!  This might actually be sufficient.  There's no actual traffic on the
connection, to my knowledge, while the PPP connection is down.  (I guess I
will have to make certain of this.)

   Date: Mon, 12 Feb 1996 00:35:31 -0800
   From: John Nemeth <jnemeth@cue.bc.ca>

	In this case, you should look at using screen,
   ftp://prep.ai.mit.edu/pub/gnu/screen-3.7.1.tar.gz.  It will let you keep a
   session running across logins.

Ah again!  I didn't know about `screen'.  It looks like it has a zillion
features we don't need, but maybe they won't get in the way.  If `rlogind -n'
doesn't do it then I'll try `screen'.

   Date: Mon, 12 Feb 1996 13:03:13 -0500
   From: "Charles M. Hannum" <mycroft@NetBSD.ORG>
   
   The kernel will drop a TCP connection after a certain number of retransmission
   failures.  Look in sys/netinet/tcp_timer.c for the code which does this.
   Increasing TCP_MAXRXTSHIFT and expanding the tcp_backoff[] array would allow
   you to disconnect for a longer time.
   
Ah yet again!  Given the dire warnings I have received about the consequences
of frobbing the guts of TCP, I will save this as a last resort; but it's nice
to know where to look if I want to give it a try.

Thanks to all!

-- Scott