My resaerch group used to have a dozen or so pmaxes.  I never noticed
any problems with the lance driver through the initial 1.3 release,
excluding those rare times when someone broke the coax chain.

My lab notebook says that a 5000/240 can sustain within 5% of media
speed indefinitely. We decommissioned the pmaxes I used to have in a
lab recently, for space reasons, so I cant easily duplicate that now.


But the two machines I have at home (I have more, but only two powered
on at any one time) work fine, _except_ when there are problems with
the repeater tree between me and the Stanford backbone.  That affects
all the machines I have at home, both NetBSD and otherwise (an IPX
running sunos 4.1.4 and my room-mate's win95 laptop.)  This is _not_ a
NetBSd-related problem, it's a local network config problem which goes
back 5 years, probably due to an interbuilding coax run grounded at
both ends, or some such.  But even there, I get "excessive collision"
messages from the Lance driver, along with a TDR value for where the
(late?) collissions occured.

In my experience, "lost carrier" usually means there's something wrong
with the cable plant, repeaters, or termination, if that helps any.


Jorgen Perhson's problem with a 5000/200 and a repeater (as a 10base2
to 10baseT converter) may be due to terminating one end of the coax
twice: some DEC repeaters terminate teh coax segment internally.  The
8-port DEMPRs do that, for sure.  If you terminate both ends of the
coax to a DEMPR, you will lose.  I don't know about the DEREPs or the
smaller two-port AUI to BNC repeaters.

Doing a ttcp between two machines with "poor performance" and
recording a tcpdump trace on a third machine would be most
informative.

Trevor Schroeder's problem I'm not sure about, and the hp300 problems
I have no clue.  I think there are some bugs in the TCP dynamics in
all NetBSD 1.3.x releases (different bugs in each release), but I dont
recall what they are, off-hand.