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

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

Sorry for the out-of-order response here.  

I think the reasons that various people favor shared trees in core
networks is a "belief" that there are significant savings in router state
and state maintenance.
We did a fairly detailed and (I believe) realistic analysis of these costs
in the EXPRESS paper and came up with numbers that totally contradict this
belief.
  I would really appreciate getting critical feedback on where our analysis
is off-track, and if not, help in dispelling the belief in state as a
benefit of shared trees.  Many in the community have had lots of brutal
experience with limited memory in routers and router vendors charging
high prices for memory, but let's not confuse the past with the future.
Memory and processing power is dramatically cheaper and getting more so.
Is anyone aware of an analysis of what shared tree state would save over
using single-source trees, using current memory prices and some credible
application?  I bet it would less than the coffee budget at most ISPs,
even if they are tea drinkers!

DRC

At 10:13 AM 5/6/99 -0400, Zheng Wang wrote:
>
>
>"David R. Cheriton" wrote:
> 
>> Anyone got an application that ISPs and router vendors would find
compelling
>> as motivaing bi-directional shared trees?
> 
>I dont believe that there are few cases that have to use one or the other.
>
>It seems to me that per-source tree and shared tree are two obivious
>approaches. The shared tree allows one to aggregate multiple trees
>into one at the expense of less optimal topology (but we cannot
>have optimal tree anyway since the unicast routing only has shortest
>path "to-destination" not "from-source").
>
>ISPs seem to like the shared tree approach in the core networks.
>
>Cheers
>Zheng
>
>-- 
>Zheng Wang
>Bell Labs Lucent Technologies
>http://www.bell-labs.com/user/zhwang
>