Cisco bug

[Robert Olson]

At 08:14 AM 6/23/99 -0400, John Connolly wrote:
>One of my local networking experts asked me about the multicast bug
>in the Cisco router at Argonne that ruined the demonstrations last
>week. 

Well, it didn't ruin any demonstrations -- it led to some hitches in the
debugging for the cruise until we put a workaround in place.

> I said I didn't know anything except that Rick Stevens thinks
>it's software and Mark Hereld thinks it's hardware.  Is there a
>place you can point us to, to find out more about what the actual
>problem is?  And is anyone thinking about a work-around in case
>"beating on Cisco" doesn't fix the problem?

The problem is related to the Cisco Catalyst ethernet switching
infrastructure we have in place here. My impression is that the basis of
the problem is that the switch(es) get in a state where multicast traffic
will be flooded to all ports; and that the emptying of the cgmp (Cisco's
multicast routing/switching tech, cf
http://www.cisco.com/warp/public/cc/cisco/mkt/gen/tech/cgmp_wp.htm) table
is related to this and somehow relates to the drying up of multicast
traffic during these episodes. 

The manifestation of this that we saw last week was periodic freezing of
video and cutout of audio.

We worked around this 	by using the Fermi multisession bridge for the
ANL-local sessions instead of the local multicast. Note that the Fermi
bridge is what we're using currently to connect the remote sessions until
wide-area multicast is debugged.

As far as beating on Cisco is concerned, one of our network folks is
working closely with a Cisco engineer and the two are deep into debugging
this. This is a problem that has cropped up elsewhere and I hear that they
may have an answer soon.

--bob