On Jun 18, Allen Briggs wrote
> I've got two NetBSD/i386 systems--one running -current as of a couple of
> days ago (+ bouyer's IDE patches), one running 1.3.1.  I was trying to
> ftp get on the 1.3.1 machine, pulling from the -current machine.  Both
> systems have DEC ethernet controllers (both AF, I think) running at
> 100Mbit, half-duplex.  They're attached to a hub with an uplink to a
> switch and one other (relatively low-trafficed) host on the hub.
> The ftp was stalling.  I said to myself--"This isn't right.  I should
> run tcpdump to see what, if anything, is transpiring."  So I ran
> 'tcpdump "host hostA and host hostB"'.  As soon as I did that, I started
> getting something like 1-2MB/sec.  Still not fabulous, but respectable.
> Since this was a relatively huge file (1.2G), I stopped tcpdump and
> restarted it a couple of times.  Whenever I stopped tcpdump, the ftp would
> slow way down.  Whenever I restarted tcpdump, the ftp would proceed at a
> more acceptable pace again.
> 

A bug in the tulip chip/driver, prventing it to receive some packets
if not in promiscious mode ? 
Would some other operartion ('ifconfig de0 down; ifconfig de0 up' or
a ping) have had any effect ?
There seems to be a problem with the tulip driver and media autodetect.
It will hang if the link goes down for a short period of time.
A 'ifconfig de0 down; ifconfig de0 up' wake it up. Also, hardcoding the
media type in ifconfig.de0 seems to avoid the problem ...

--
Manuel Bouyer, LIP6, Universite Paris VI.           Manuel.Bouyer@lip6.fr
--