Re: Simple Multicast - building a case for a BOF or WG

["David R. Cheriton" <cheriton@DSG.Stanford.EDU>]

Ken,
  I wonder how much of an issue the QoS/R will ever be in practice because
we seem to have evolved to a pretty hierarchical structure to internetworks,
so the best QoS will typically come from following the hierarchy, i.e.
from source thru tail circuit to local ISP to wide-area ISP, and then out
and down to receivers.  Do you have a network topology of concern where
this does not work out?
  Also, I dont believe that single-source has to be constrained to just
use shortest path, although I agree that would require coordination with
the RPF check.  It might be interesting to explore support for asymmetric
routing as part of this issue, if we believed that satellites had a future.

DRC

At 08:57 AM 5/6/99 -0400, Ken Carlberg wrote:
>
>> On this latter point, and following up on Adam's query on costs, I'd like
>> to understand the compelling application(s?) that motivate the
bi-directional
>> shared trees.
>
>i don't believe there are any well-known apps, ones that we see on the
>mbone and that get discussed on different multicast lists, that would
>break if they couldn't use bi-directional trees.  
>
>one can argue that DIS-type applications operate a bit better with bi-
>directional trees, but they will still perform with uni-directional
>trees (at least given today's number in groups, participants, and 
>topology).  
>
>outside of these 'normal' sets of applications, one thing that is 
>difficult to do with uni-directional trees that use RPF checks on 
>the source, is to try and tweak these tree construction designs to 
>support QoS or Quality of Route (QoR).  as an example, trying to 
>promote high fan-out towards the leaves of a tree is much easier
>to do with bi-directional trees, where i can graft a branch onto 
>any on-tree node, as opposed to constraining the graft process to
>only the shortest path towards the core.
>
>-ken
>
>
>
>
>