On Fri, 15 Sep 2000 22:57:05 +0930 (CST) (06:27 PDT)
        blymn@baesystems.com.au (Brett Lymn) wrote:

>According to Ignatios Souvatzis:
>>
>>Thats not pppd that gets blocked. Thats your modem -- tty device -- connection
>>that fails to transmit the "hangup" signal.
>>
>
>Sorry, no.  I am not sure what it is but I can make my ppp link stop
>just by remote displaying emacs over the link, the modem definitely
>stays off-hook so the line does not go away underneath me and this
>happens if I have dialled into a Cisco modem server or in via an annex
>terminal server (don't care what other people say the annex has worked
>well for me for _years_).  I too have found that a constant ping, or
>just about any dribble of traffic seems to help keep things going for
>longer but the link will still, eventually, lock up.
>
>  The behaviour seems to be related to the link usage.  I normally
>forward X over ssh to get compression & security.  I sometimes have
>had to forward the FW-1 management tool display across my ppp link,
>this works fine BUT if I try to forward emacs over the same link it is
>pretty much guaranteed the link will lock up before I have finished an
>email reply - and, yes, it is the link locked up because pings fail
>and my other sessions are dead in the water.

I have some seemingly similar behavior with no modem, tunneling pppd 
via SSH.  My configuration looks like:

  +---------+              +---------+
  |         |              |         |
  |  NetBSD +--------------+  NetBSD |
  |    F    |              |    H    |
  +----+----+              +---------+
       |
       |
    PPPD-SSH  (DSL or Cable)
     tunnel
       |
       |
  +----+----+
  |         |
  | Solaris +-----+-------------------------------+
  |    W    |     |                               |
  +---------+     |                               |
                  |                               |
             +----+----+                     +----+----+
             |         |                     |         |
             | Solaris |                     |  HP-UX  |
             |    X    |                     |    Y    |
             +---------+                     +---------+

I'm using ppp-2.3.11 (McKerras) on the Solaris-W machine.

If I:

    rlogin Solaris-W
or
    rlogin HP-UX-Y

from my home machine (NetBSD-H) the session is stable and stays
up as long as the tunneled pppd connection is viable and, in fact,
many times have survived the pppd/ssh connection going down and
coming back up.

If I:

    rlogin Solaris-X
or
    rlogin Solaris-W
    and then rlogin Solaris-X

I can hang the session pretty easily.  These sessions seem to be very
unstable.  It is very easy to hang the session with an ls that produces
more than one screen of data or a "more" with just about any amount of
data.  If I break out of the session with ~., so that I'm back on
Solaris-W, that session seems stable and continues to work fine.  I can

    rlogin Solaris-X
or
    rlogin HP-UX-Y

and get the same results over and over (Solaris seems flakey, HPUX
seems stable).

If I'm directly on Solaris W (using the directly connected monitor)
connections to Solaris-X and HP-UX-Y seem very solid.

Solaris-W ipnat's the NetBSD-H address to a logical address (hme0:1)
on Solaris-W.

I'll gladly post any relevant stuff.  There is so much that it could
be, I hesitate to dump it all in here to start.  The path is common
until it is sent out on the wire to X & Y.

I'm suspicious that it is some setting on X, but I'm pretty baffled
about this and have nary a clue where to start looking.

Paul Newhouse
piminx@home.com
newhouse@rockhead.com