>>>>> "dy" == David Young <dyoung%pobox.com@localhost> writes:
dy> (On the DHCP server, TCPDump only showed the fourth request.)
this might be the problem:
http://mail-index.netbsd.org/port-alpha/2009/04/21/msg000263.html
the short version is: if you are plugging into a managed switch like
Cisco or Extreme, configure the switchport connected to the netbooting
machine as an edge port. ('conf t / int <whatever> / spanning-tree portfast'
or 'conf stpd <name> ports link-type edge <m:n>')
If you can't do that, plug in a crappy SOHO switch between the managed
switch and the netbooting machine, which will hide the link-beat
up/down's from the managed switch. This works because crappy switches
are usually less obstinate and use something like the automatic
edgeport detection state machine in 802.1w, or if they are really
crappy don't have an STP at all.
If you do both you may have problems: some managed switches will
report ``edgeport inconsistency'' and disable the port if they see an
802.1d/802.1w-speaker on what they were told should be an edge port.
I'm not sure the exact rules for this but have seen both Cisco and
Extreme do it.
It may be a particular problem for NetBSD because NetBSD's stage2 is
turning the card on and off, or cycling through speeds/duplexes, or
whatever it's doing, on some timer that conflicts with STP so the port
never transitions to forwarding. Also the stage2 should probably try
forever like Sun OB does, and should be certain to discard ALL
information between one failed try and the next (unlike Etherboot :).
Attachment:
pgp_f_9VvdK86.pgp
Description: PGP signature