All things related to multicast

[Lea Roberts <>]

greetings wg-multicast folk -

So it looks like the multicast issue between the University of Michigan
and Stanford is the unfortunate interaction between a possible Juniper
bug and an older Cisco "feature".

During lunch at Joint Techs this Tuesday, 7-Feb-06, a bunch of us got
together to examine in more detail the problem reported by Steve Senger
here at Joint Techs on Monday regarding multicast join latency between
the University of Michigan and Stanford University.  Present were Bert
Rossi and Laurie Kirchmeier from Merit Networks, Steve Senger from
University of Wisconsin - LaCrosse and myself, from Stanford.

What we discovered was that the delay was in the Stanford network
(again...  :-) <sigh>
Specifically, the MSDP SA appeared almost immediately in the Stanford
border router, which connects to CENIC, but it did not appear at the
Stanford RP (one hop away) until ~40 seconds later.  Once the SA arrived
at the RP, the join propagated correctly in a timely manner.   Steve
remembered that the previous bug in the CENIC router code had to do with
the forwarding of MSDP SAs with encapsulated data, so we proceeded to
try to determine if the Michigan RP was encapsulating data in the MSDP
SA.  Also, Laurie tried sourcing from a host at the I2 offices in
Washington DC and we discovered that it also experienced the join latency
from Stanford.  Meanwhile, Bert found that the RP in Michigan, a
Juniper, was configured not to encapsulate.  He changed the setting but
the delay was still there.  At that point, we decided to turn on MSDP
debugging in the Stanford RP.  We then determined that the MSDP SA from
U of Mich still did not contain encapsulated data and neither did the SA
from I2 in DC.  Steve then verified that joins from Stanford to
Wisconsin were still occuring in a timely manner and we observed that
the Wisconsin RP, also a Juniper, was sending encapsulated data with the
first MSDP SA.

So the problem seems to be the delayed forwarding of MSDP SAs by the
Stanford border router, a Cisco 6503 with a sup720 running IOS version
12.2(17a)SX4.  The preferred fix will be to find and upgrade the IOS
to a version that forwards all MSDP SAs in a timely manner whether or
not they contain encapsulated data.  Another Michigan-specific fix
will be to convince the RP to encapsulate.  It seems there may be a
bug in the latest JunOS which the Michigan people will be pursuing,
trying to find out why it wasn't encapsulating after the configuration
was changed to enable it.

Never a dull moment in multicast land!                  /Lea