Re: New Internet Draft on automatic (end-user) tunneling for SSM

[Hugh LaMaster <lamaster@nren.nasa.gov>]

On Thu, 1 Mar 2001, Marshall Eubanks wrote:

> My personal feeling, based on talking with a lot of service providers, is that
                                                      ^^^^^^^^^^^^^^^^^
I'm not sure if you are talking about Tier-1 or Tier-2/3 service providers.
I would guess that some Tier-1 providers will do this "for free" as long
as they could be sure that it would have near-zero impact on their
core/edge routers, as stated previously.  To me, that means that that
should be a way for the protocols to allow all control traffic to be
forwarded directly to the helper router without CPU intervention--
as discussed previously, when you are forwarding at OC-12/OC-48/OC-192
speeds, you need to maintain pretty strict separation of
routing processes and the switching functionality in the router.
(MSDP bursty-source encapsulation is something that violates this,
of course, so I would like to see alternatives proposed for that.)

The Tier-2/3 providers are in a different economic situation.  
So, I'm assuming that you are implying Tier-2/3 here.

> they will only provide tunnel relays if they are paid to do it. The unicast
> traffic that results from tunneling, of course, they will carry without
> complaint.

I wonder if the Tier-2/3 providers will see it that way if they
ken how much tunnel traffic is being shoved at them.  Of course,
the traffic from N streaming video streams is the same, but,
I'm worried about the knowledge gap-- they may see it as different,
even if it isn't really.

> I still do not see how auto tunnels will scale. If my tunnel relay's will only accept a fan-out of 2,
> then to serve 2048 people will require 10 generations of tunnels (and about 1000  
> tunnel servers). Even
> if I accept a fan-out of 5, 2048 people will require ~ 5 generations.

I don't think it *will* scale.  That is impossible, actually.
But, it might scale up just well enough to provide an incentive 
for Tier-2/3 providers to turn on multicast.  I'm agnostic
on that particular point.  I think the fanout you can get 
from a mid-size router can be much more that you state, though.
I know that you can do replication for 10-20 tunnels on a 
midsize software-forwarded router -- the limiting factor is 
the total outbound traffic -- that is, how large are the 
streams?  Once this is up and running, there would be an 
incentive for the downstream Tier-2/3 provider to turn on 
multicast and save 20-30 Mbits/s traffic incoming.  The 
incentive for the Tier-1 provider would be that by intercepting 
the requests at the edge, they would save the same amount 
of traffic over their backbones, so they get that immediate
benefit, plus, the long-term benefit of eventually having
the Tier-2/3 provider turn on native multicast.

> This seems like an awful long chain (with accumulation of jitter, 
                                          packet loss, etc.) and a
> lot of coordination for an actual, workable, product.(Yes, this 
                                          is what the routers
> do for ISM or SSM, but they are designed to do it. 

                                They also tend to cost a lot more than
> I suspect people want to spend for tunnel relay's.)