[ppml] whither 2005-1 (was: Policy without consensus? )

[Marshall Eubanks]

Hello;

On Jan 23, 2006, at 11:34 PM, Paul Vixie wrote:

> # I am an IETF oriented folk myself, and I have been down this road  
> before.
> # This, my comments re Multicast. We spent way too much time  
> worrying about
> # problems of success; as (relative) success is coming, it isn't  
> coming in the
> # way that we worried about.
> #
> # Just an observation.
>
> one way of interpreting these remarks is "multicast is still not in  
> wide
> interdomain use because its creators held it back until they could  
> figure
> out a deployable model".
>
> another is "multicast is still not in wide interdomain use today  
> because
> its creators never found a deployable model."
>
> i could agree, and simultaneously disagree, with either  
> interpretation.
> but i'm wondering-- marshall, what exactly do you mean here?
>

Neither, exactly. I would not agree with the former, and the latter  
is not entirely true.

A lot of cycles were spun worrying about things that never happened.  
That was just my point.

If you want my opinion, here it is :

Multicasting, round 1, (DVMRP MBone) never found wide interdomain use  
because it just
wasn't suitable. In some sense, though, I think it came closest to  
universal deployment. And it made
Mark Cuban $ 6 billion USD or so.

Multicasting, round 2 (PIM ASM) didn't find wide interdomain use  
because the service model
has problems (ASM's and MSDP's sensitivity to interference and DOS  
attacks, and difficulty in actually
getting it to work reliably interdomain). The timing was really bad  
too (just late enough that
the money from the dot com boom primarily went to unicast streaming).

Multicasting, round 3 (SSM and IGMP v3) . Now we are seeing lots of  
"walled garden" deployment of multicast.
It's mostly ASM. It will move into SSM in the next few years. My  
personal feeling is that the walls of the gardens will eventually be  
breached by the implications of Zipf's law, but that remains to be seen.

Note that SSM was finalized in Adelaide, just about 6 years ago. And  
it is still not universally available !
(IGMPv3 support not in every OS, not in most applications, and not in  
most IGMP snooping switches). I would guess than there are ~3 years  
to go there before it can just be assumed. I see no reason why SSM /  
MBGP can't be generally deployed, but it is taking a long time.

It takes a long time to change things at the routing layer in the  
21st century. Be careful what you put into millions of boxes, as it  
will take a long time to change it.

> if one has to choose among types of disasters, we would mostly all  
> prefer
> a success disaster to a failure disaster.  but what if we won't WANT a
> disaster?

I have come to realize that new protocols have a certain amount of  
political capital. The balance of multicast is pretty depleted at  
present, even though usage is accelerating. Going from IGMPv2 to v3  
in the way that was done was a big hit IMHO because it required both  
OS and edge box upgrades. A lot of the work on BGMP (not MBGP) and the
related MADCAP address allocation stuff (deployed in Windows and  
never used) also burned up political capital with no real benefit. We  
had better not need to change anything in multicast now, as it will  
be real hard
to get anything new into OS's and hardware IMHO. Fortunately, I don't  
see anything needed at present.

What I see in IPv6 is a tiny amount of actual deployment  
(considerably smaller than inter-domain multicast), with a strong  
possibility that anything that really needs it will be a walled  
garden (the IPv6 as a super kind of NAT idea). If small entities want  
to multihome, I think that should be encouraged.

Regards
Marshall



> _______________________________________________
> PPML mailing list
> PPML at arin.net
> http://lists.arin.net/mailman/listinfo/ppml