Great thanks to all who responded: Nikola Milutinovic, John Lanier, Shawn Cromer, Trevor Osatchuk, Martin Roende Andersen, Rochelle Lauer, Bluejay Adametz, Colin Bull, Bård Tesaker, Chris Ruhnke, Iain Barker, Jenny Butler and Michael Wheelock.
Chris brings it to the point by saying
"Concensus opinion is to hard configure both the adapter and the hub to use desired speed and duplex".
Bård clarified it further and gave some more useful information:
"The connected switchport is probably set to auto. Contrary to speed, the duplex setting of the peer cannot be detected. A special "auto negotiation" protocol is used to agree on duplex. Set your interface to "AUTO" to enable this protocol or else the switch assumes half duplex. Alternatively you may force the switchport to use full duplex if the switch is managed. Some switches gives a 30sek hang on link up (typicaly cisco) due to STP spanning tree evaluation. This is particulary annoying on tulip since it does a link down/up on config changes (pfconfig ao). Set the switchport to portfast to avoid this."
Trevor added:
"Conventional wisdom will tell you not to do this [...] everyone has said not to use auto-negotiate, but it was the only thing that worked."
Indeed, I have heard from HP support before that "autosensing does not work properly with the tulip", they recommended to hard set it to FastFD. The man page for tu(7) also says as much, although it does say something about DE500-AA and DE500-BA apparently supporting auto-negotiation, however it says nothing about the newer cards that we have.
I was able to use our cards (21143 chipset rev 4.1) with the auto-sensing Cisco catalyst switch simply by setting the card to "Auto" at SRM level.
So, it seems all works well as long as both sides are auto-negotiating or set to the same fixed speed and duplex.
Thanks to everyone for taking your time to help!
Michael
Received on Wed May 05 2004 - 14:10:15 NZST