[4122] in linux-net channel archive
Re: UDB Ethernet problem, part 2
daemon@ATHENA.MIT.EDU (Michael O'Reilly)
Tue Aug 20 23:30:38 1996
In-reply-to: "Joshua M. Thompson"'s message of Thu, 15 Aug 1996 19:14:58 -0400 (EDT)
To: "Joshua M. Thompson" <invid@optera.com>
cc: linux-net@vger.rutgers.edu
Date: Wed, 21 Aug 1996 10:19:08 +0800
From: "Michael O'Reilly" <michael@metal.iinet.net.au>
>>>>> ""Joshua" == "Joshua M Thompson" <invid@optera.com> writes:
> (I'm sending this to linux-net as well now, because this is a _very_
> strange networking problem)
> Quick recap: DEC Alpha Universal Desktop box, running RedHat 3.0.3
> but with kernel 2.0.12 with DE4x5 driver (also used to run 2.0.11
> with Tulip driver, with same results). The problem is that at random
> times, my Ethernet connection (UTP) goes "dead".
> The last time this happened, I did some experimentation and
> discovered some interesting facts about this problem :
[ .. ]
> 4. If I leave a ping running (not ping -f, just plain ping) and then
> start doing the plug/unplug thing with the Ethernet cable, then when
> it _does_ start working I get _all_ my ping respones. The first 10
> or 20 have ping times on the order of tens of seconds (10,000s of
> ms). To me this seems to indicate the DEC _did_ receive all those
> packets, but it just buffered them until something I did "unclogged"
> it. If it weren't for this strange behavior I would just go with the
> assumption that my DEC is broken.
No, It indicates that the sending host buffered them waiting for an
ARP reply.
I'm using a PPro 200, 128 meg ram, SMC ethernet card running
2.0.9. Spot the common variable. :)
If I have another machine do proxy arp for the machine with the SMC,
it's flawless. The problems seems to be simply that the SMC doesn't
always respond to ARP requests. In fact, I'm not sure it ever does.
My workaround was simply to do
arp -s <ip#> <mac address of SMC> pub
on another machine, (well, actually the equivlant on the cisco), and
voilo! all is well.
I've NO idea why the SMC doesn't respond. I don't know if it's no
responding because it's not seeing the arp request broadcast, or if
it's just no seeing that it's a request for it's ip#.
Actually, my guess is it's simply not seeing the arp request, which
indicates a problem with broadcast packets.
Hmm. Quick test. a-hah!! If I ping 255.255.255.255 from the cisco,
everything EXCEPT the machine with the SMC responds. That's the
answer. The SMC isn't seeing ethernet broadcast frames.
Anyone got any guesses on why? (ps. my SMC is an SMC EtherPower
based on the 21041 chip)
Michael.
