[4049] in linux-net channel archive
strange inter-linux ethernet problem.
daemon@ATHENA.MIT.EDU (Paul Levin)
Fri Aug 16 12:58:35 1996
To: linux-net@vger.rutgers.edu
Date: Fri, 16 Aug 1996 07:29:43 -0400
From: Paul Levin <Paul_Levin@postbox.ius.cs.cmu.edu>
Howdy y'all....
I've just subscribed to this mailing list, so perhaps my question
has been previously answered; my regrets if this is the case.
I have two linux boxes hooked up to my ethernet connection, although
the plan is to have them run on a very small network (three
linx boxen only). My problem is that they do not communicate
well between them although each can communicate with
other (unix) machines on my current network.
My configuration is:
LB1 (linux box 1)--pc with pentium triton chip, running
linux 1.3.85, all ethernet options enabled, slackware
ethernet daemons (including wu-ftp-2.4, and NetKit-A-06
software); 3com 3c590 combo.
LB2--pc with passive backplane configuration, neptune
chip running linux 1.3.86, all ethernet options enabled,
RZ1000 bug-fix patch enabled, same ethernet suite as LB1;
same 3com card.
Symptons:
1) LB1 can ping an arbitrary unix machine (UM), itself,
and LB2 with reasonable time metrics.
2) LB2 can ping UM, itself, and LB1 with reasonable time
metrics.
While on the work-wide ethernet connection, the
following were observed:
3) LB1 can telnet into LB2 and UM; LB2 can telnet into
LB1 and UM; no timing delays observed in any of
these connections.
4) LB1 can ftp into UM with timing metrics similar to
inter unix-machine ftp; same for UM doing ftp into LB1.
5) LB2 can ftp into UM with timing metrics only a bit
slower than those observed in 4).
6) LB1 can ftp into LB2 but the timing metrics are about
a factor of 20-50 slower than those found in 4).
7) LB2 hangs on attempts to ftp into LB1 (after about 2K
bytes sent).
8) When the two machines are isolated on their
own net, LB1 hangs on attempting to telnet into
LB2 (either telnet by name or by IP number).
I ran netstat on LB1 when I try to ftp into it from LB2 and
find a very long time between closing the ftp program on
LB2 and having the connection completely severed on LB1; I also
get repeated "SOCKET DESTROY DELAYED" messages on LB1. When
I run tcpspray on LB1 with LB2 as the target, I also have
the slow timing metrics as when I ftp.
This behavior occurs even when I swap the position of the
3com card on LB2 (upstream vs. downstream of the pci bridge
chip).
Not only is this not fun, but I have a major deadline staring
me in the face (as in 'real soon now!').
Help!
Thanks in advance to everyone out there for all your suggestions.
Best regards
PDLevin, The Robotics Institute, CMU
ps. I've had *lots* of problems when I attempt to compile
network-related software, especially with apparent incon-
sistencies with the /usr/include/net and /usr/include/inet
files (either the ones I have or else the ones supplied).
I've tried repeatedly several tcpdump utilities, and can't
recompile wu.ftp. (I just upgraged to gcc 2.7.2). Has anyone
else had this problem, or am I missing something?
