Subject: AppleTalk & EEpro100 cards (was Re: [netatalk-admins] Massive dropped connections - any ideas?)
From: Keith Clarke (keithc@dcs.qmw.ac.uk)
Date: Mon Apr 26 1999 - 09:56:27 EDT
>The multicast support on the Intel and 3Com cards is crap compared to that on
>the Tulip based boards.
>
...
>Jonathan
And yet UAR works with these cards tho' atalkd does not...
We have two file servers - Netatalk 1.4b2 running under FreeBSD 2.2.6, and
Netatalk1.4b2+asun2.1.0 running under FreeBSD 2.2.8. We want to make these
services available to another subnet connected via a software router, and
thought to use atalkd for the DDP routing (atalkd from
Netatalk1.4b2+asun2.1.3).
Simon Boggis of our Dept just posted the following to the list
<mailto:linux-eepro100@beowulf.gsfc.nasa.gov> -
>I have been trying to set up appletalk routing only (no file serving) on a
>linux router box which is equipped with 3 dual port eepro100+ (82558B)
>cards. The system is linux 2.2.6/glibc 2.1.1 (Debian Potato), however I am
>using the 1.07 eepro100 drivers (rather than 1.06, as shipped with the 2.2.6
>kernel). I have been attempting to bridge appletalk between our main
>network on
>eth0 and a private subnet on eth3.
>
>Curious things happen when the appletalk module is installed -
>one gets lots of:
>
>Apr 20 18:34:34 MACHINE kernel: protocol 9b80 is buggy, dev eth0
>Apr 20 18:35:20 MACHINE kernel: protocol 0008 is buggy, dev eth0
>Apr 20 18:38:21 MACHINE kernel: protocol 0400 is buggy, dev eth0
>Apr 20 18:40:14 MACHINE kernel: protocol 3781 is buggy, dev eth0
>
>in /var/log/debug if you tcpdump eth0.
>
>When netatalk (atalkd only) gets fired up, appletalk connections hang quite
>quickly, accompanied by messages in /var/log/messages:
>
>Apr 15 17:52:36 MACHINE kernel: eth3: Transmit timed out: status 0050
>0000 at 17015/17030 command 00000000.
>Apr 15 17:52:36 MACHINE kernel: eth3: Trying to restart the transmitter...
>
>There has been a lot of discussion on the netatalk-admins list concerning
>this,
>which I have been following, and the conclusion there was that the hardware
>was broken, but could be worked around by setting multicast_filter_limit=3.
>I have no idea why '3' particularly, however in my case I found the solution:
>
>(a) didn't work (macs still hang after a minute or so) (:
>(b) stopped mrouted from working on the machine
>
>and therefore wasn't much of a solution at all for me.
>
>Having consulted a colleague, I was assured that appletalk works fine on pc
>hardware with eepro100's using Novell Netware. I therefore was suspicious of
>the diagnosis of a hardware fault in the eepro100 offered above.
>
>In order to isolate software from hardware, I tried the much older Univeral
>Appletalk Router (UAR 1.0.16 September 1995) which doesn't use the appletalk
>support in the linux kernel. This works absolutely fine. This leads me to
>believe (perhaps incorrectly) that the hardware and eepro100 driver are fine,
>but that either netatalk and/or kernel appletalk support is broken.
>
>Has anyone else been having trouble with appletalk on the eepro100? More
>importantly, does anyone have a view on whether my above analysis is correct
>(i.e. the hardware and driver are OK).
>
>Simon
>
UAR working "absolutely fine" describes a simple test - a single Quadra620
running OS8.1 on the private subnet, running an AppleScript copying a 30MB
folder (400 or so files) continuously for three days from the
AppleTalk-only server. This test caused a hang within ten minutes using
various different configurations of atalkd as the router.
Are we doing something wrong here (i.e. trying to use atalkd as a router,
with no afpd) ?
We'd be really happy to help out with debug logs etc - just tell us what's
needed. Of course a possibility is that Netware and UAR are accidentally
working round bugs in the card...
thanks
Keith Clarke
This archive was generated by hypermail 2b28 : Sat Dec 18 1999 - 16:16:38 EST