Editor's note:  Minutes received 4/14/93.  These minutes have not been edited
                and the attendee list has not been appended.

Reported by Tony Ballardie


		Inter-Domain Multicast Routing BOF Minutes

The IDMR BOF session was held on Friday, 2nd April, 1993, during the
Columbus IETF meeting. The meeting was chaired by Tony Ballardie, 
University College London. These minutes were taken by Benny Rodrig 
of RND - RAD Network Devices Ltd.

The CBT protocol was discussed for the duration of the session, in 
particular, the changes in the protocol since last November's IDMR BOF
session.

One possible solution to multicast scope control was presented, based
on having a separate group per level of scope required. This resulted
in a considerable debate as to how multicast scoping should be defined
and the requirements of users it should be able to satisfy. The solution
presented was deemed unsuitable, and it was agreed to continue the
the discussion on the idmr mailing list. The conclusion was that the 
group should work towards a concise definition of multicast scope control.

Part of the group initiation procedure includes the group initiator (host)
updating DNS (via a system administrator) with the relevant group information. 
The concern was raised that the procedure is likely to be too slow. 
This part of group initiation may need to be revised to keep group 
join latency to a minimum.

There was a brief discussion on the issue of multicast data packets
carrying the group-id as an IP option. The conclusion however, was that
there was no more suitable alternative.

When a parent/parent link fails, it was decided that both the flush-tree and
re-join mechanisms should be implemented. Which of the two mechanisms
eventually becomes redundant can only be decided after a period of 
performance testing.

It was decided that non-primary cores should be less stringent in accepting
join-requests. Also, an additional error detection mechanism is required
by routers to distinguish on-tree packets arriving via a child as 
link-level unicast.

The draft includes no mention of child-parent "keepalives", and no
mention of a handshake protocol between a group initiator and 
potential cores for that group. The draft needs updating in these respects.

The control packet format should contain a group-id TYPE as an additional
protocol independence feature.

Finally, Paul Tsuchiya gave a brief description of how CBT will run over
Pip.