Ethernet Woes

[Peter W. Flur]

We are experiencing a very interesting problem with a couple of our
machines here at Georgia Tech.  Two of our machines are losing between
20-30% of their packets when going through a router onto the campus 
network.

Now for the details.  My floor has its own subnet separated from the 
campus network by a PC router with two WD8003 ethernet cards 
running PC-ROUTE.  Their are approximately 25 machines on the subnet
including PC's, MAC's, NeXT's, Sun's, DEC's, an AT&T 3B2-1000/80, and
an Intel IPSC/2 Hypercube.   When communicating between two machines
within the subnet, everything is fine, and no packets are lost, independent
of which two machines are being used.  When any PC, MAC, NeXT, Sun, or
DEC machines talks with a machine outside the subnet (using ping, telnet,
ftp, etc.) everything works fine and not a single packet is lost.  

However, the problem comes when either the 3B2 or the IPSC try to communicate
outside the subnet.  Each exhibits a 20-30% packet loss very consistently.
This is true whether it is ping'ing or it is being ping'ed.  

We have concluded several things:
1) Only the two System 5 machines are having problems
2) The two machines work fine within the subnet
3) No other machine (all BSD) has problems communicating with the outside world
4) The netmask and broadcasts ARE set properly
5) The number of STREAMS and QUEUES appear to be more than sufficient
6) The cables and connectors to the two machines are fine

After putting a sniffer on the subnet, we have found that the echo packets
are being sent out fine, and the replies are sometimes heard on the subnet
even though they appear lost.  The only other significant detail that we
have found was that the "time to die" for packets to/from other (BSD) machines
appear in the range 200-300.  The packets from the two Sys5 machines appear
to be in the range 15-30.  This is theoretically the maximum number of 
hops remaining before death, and should be well within range.  

Can anyone offer any good advice?  Has anyone seen similar problems?  Your
help would be greatly appreciated.

Peter
-----------------------------------------------------------------------
		   Peter Flur, Research Engineer II
		   Georgia Institute of Technology
      School of Electrical Engineering, Atlanta, GA  30332-0250
		     E-MAIL: flur at duke.gatech.edu
			PHONE: (404) 853-9355