🚀 go-pugleaf

First: the problem does not occur with NDIS or ODI drivers.

We discovered that SMC Ultra cards with the packetdriver PKT8000 
sometimes sends out a packet which it has just before received itself.
We can't prove the card is retransmitting ethernetpackets that were 
destinated for the card but we can prove the card is retransmitting 
broadcastpackets. In fact we can easily reproduce this behaviour. 
Not the whole packet is retransmitted, only the first 64 bytes. That 
means the card is sending out packets with a source address that is not 
the MAC address of the card itself but the address of the sender of the 
original broadcast packet. That means too that a new CRC is calculated, 
based on the truncated packet. In ethernet terms it's a valid packet.
If you have Novell servers complaining about incomplete IPX packets, 
loopback errors or <ethernetaddress> is using our <IP address>, than you are 
dealing with this problem. In our case with 28 Novell servers, 
96 printservers and more than 300 Ultra cards, we see every minute a 
duplicated IPX packet. Most of these packets are SAP and RIP packets send 
out by Novell servers and printservers.
ARP requests are also duplicated but are harder to find.

The problem is dropped to SMC almost a year ago and the last news 
(half year ago) was that the problem was delegated to Russell Nelson.

Because this misbehaviour seemed to be fairly innocent, except for 
the log files of the Novell servers, we didn't take further action.
Untill !!!! Last months most of our Novell servers were reconnected 
to ethernetswitches (ironically of SMC). Each server has now it's own 
port. These switches are of course self learning. 
When a Novell server is sending out a broadcast and that packet is 
duplicated elsewhere in the network, than the MAC address of the Novell 
server will move to another port of the switch and the Novell server will 
become unreachable. We were triggered because our Spectrum 
networkmanagementsystem (of Cabletron) lost contact with Novell
servers several times a day. We measured then that indeed a Novell 
server can be up to 18 seconds unreachable. The MAC address of the Novell 
server in the routing table of the switch moves back as soon as the 
Novell server sends out a packet.

The intention of this posting is first to warn you and secondly to 
put more pressure on SMC to solve this problem.

More information (test setup, HP analyser programs and output) is available
for anyone who is interested.


M.A.G. van der Walle
Datacomm. and Networking

Tiburg University
The Netherlands




---------------------------------------------------------------------
M.A.G. van der Walle            | e-mail : M.A.G.vanderWalle@KUB.NL
Datacomm. and Networking        | tel    : +31  13 662091
Tilburg University              |
---------------------------------------------------------------------