eepro100 Transmit timeouts - time to cut my losses?

Osma Ahvenlampi oa@razorfish.fi
Wed Sep 15 02:26:45 1999


Scott Tyson <tysons@deepwell.com> writes:
> options eepro100 multicast_filter_limit=3
> 
> What does this do exactly and besides fixing a possible crash what are
> you loosing in functionality?

The Intel chip has a built-in 64 entry hardware multicast address
filter. However, configuring more than 3 addresses on the address list 
requires an extended configuration frame, and the way the eepro100
driver passes that frame to the chip seems to have some kind of race
condition which might cause a crash if another (multicast?) packet is
received at the same time. That's my theory, anyway.

The option will limit the number of multicast addresses configured
onto the hardware filter - after the limit is reached, the card will
be put into promiscuous mode if any more multicast streams need to be
received. That will cause a slight (as in I haven't even been able to
measure it on a switched network) degradation in performance.

This is pure theory, I haven't seen the chip specs and only partially
understand what the driver is doing. However, it has always fixed the
problem for me, and I'm not the only one. I have heard of cases where
it hasn't helped, though, so you might be out of luck.

Oh, and you might have multicast traffic on your network without
knowing about it. AppleTalk (EtherTalk) uses multicast, and so does
NTP and some routing protocols. Then there's of course MBONE and other 
IP-multicast applications that you probably know about if you're
playing with them.

-- 
|    ||   |||  ||         r a z o r f i s h , helsinki
                          tel +358-9-612-23460
osma ahvenlampi           fax +358-9-612-2345
                          http://www.razorfish.fi