Subject: Re: [netatalk-admins] AppleTalk crashes EEPro100 on a bonded etherchannel
From: Magnus Stenman (stone@hkust.se)
Date: Wed Nov 25 1998 - 07:52:20 EST
Osma Ahvenlampi wrote:
>
> Dejan Ilic <svedja@lysator.liu.se> writes:
> > *) 3Com hardware support sucks. Multicasting is "1-bit" selected.
> > Either on or off, no hardware filtering at all. You receive ALL
> > multicasting trafic if you have it all.
>
> Yep. All multicast filtering is done in software with all 3Com cards.
>
> > *) Intel networking is better, something like it is able to separate a
> > couple multicasting-channels in hardware, something like "3-bit" I
> > think it was.
>
> EEPro100 has a 64-address hardware multicase filter. However, the
> Linux driver still has a problem that it crashes if it receives
> traffic while being configured for >3 multicast addresses (the first
> three fit into the first configuration frame, while the rest need an
> extension frame). Workaround:
>
> options eepro100 multicast_filter_limit=3
>
> into /etc/conf.modules (or if your driver is built into the kernel,
> the same option to the kernel).
>
> > *) Their favorite is Digital Tulip-based cards (Dec-21something) that
> > can be found in cheap cards from DLINK for example. If I remember
> > correctly it is magnitude better in handling multicasting trafic
> > compared both to 3com and Intel.
>
> Other Tulip cards are SMC EtherPowers. Tulip has a 512-address
> hardware multicast hash filter and 16-address perfect filter. I have
> no personal experience with Tulip cards, so I can't say if the driver
> works well.
I use D-LINKs (DFE-500TX) with tulip chips;
tulip.c:v0.89H 5/23/98 becker@cesdis.gsfc.nasa.gov
eth0: Digital DS21140 Tulip at 0x6c00, 00 80 c8 45 dc aa, IRQ 11.
And they work very good, both in 10 and 100Mbit
(but the Macs do not run very much faster with 100Mbit...)
/magnus
This archive was generated by hypermail 2b28 : Sat Dec 18 1999 - 16:33:46 EST