[AusNOG] NBN Co Layer 2 vs. Layer 3 Services

[Mark Smith]

On Thu, 22 Apr 2010 17:41:00 +1000
Paul Brooks <pbrooks-ausnog at layer10.com.au> wrote:

> On 22/04/2010 5:11 PM, Daniel Hooper wrote:
> >
> > "but will be forced to travel all the way out to an interconnect point 
> > and through RSP1, then be handed over to RSP2"
> >
> > That's just the nature of the beast, if factory #1 was really serious 
> > about exchanging large volumes of traffic to the premises next door 
> > wouldn't they be eligible to purchase a layer2 style of service from 
> > an RSP? I haven't seen any thing so far that mandates that providers 
> > must only sell layer3 services. Surely the RSP could apply for a VLAN 
> > to be mapped between port X on customers ONT and mapped to another 
> > port on another ONT?
> >
> 
> NBNCo won't sell a VLAN directly between two ONT ports to build an 
> end-to-end service. NBNCo's services are all  ONTport-to-POI.
> 
> But I agree thats the nature of the beast, its exactly what happens now 
> when two locations are connected to two different providers - or even to 
> the same provider. If factory#1 was serious about large data volumes to 
> a location nearby they would find a local carrier to dig some fibre or 
> establish a microwave link directly between the two buildings, and 
> bypass the NBN connection. Just because a building is connected to the 
> NBN doesn't mean thats ALL they can be connected to.
> 
> The argument is an example to highlight a concept of "efficiency". My 
> belief is that strict efficiency in terms of shortest-geographic-path is 
> a concept that is largely out the window and immaterial. We don't care 
> about efficient routing of telephone calls these days, its all about 
> least-cost not shortest-path - I can call the person next door and have 
> the call routed around the planet a couple of times, and I don't care as 
> long as the audio quality is OK.  With ubiquitous fibre infrastructure 
> leading to almost unlimited bandwidth in practical terms, the only 
> effect from non-shortest-path traffic flows that the end-user expriences 
> is a few millisecs of latency - if this causes capacity issues for the 
> provider, than the provider has to find a way around it by finding a 
> more optimal path, or possibly finding a longer but cheaper path so they 
> can afford more capacity - either way, the argument that  traffic 
> between two locations that are next door to each other shouldn't ever 
> flow beyond the end of the street doesn't hold in my view.
> 

I think it's the fantasy of a full mesh network that traps people.

The ultimate uncongested and lowest latency network is a full mesh one
- right from the back of your PC to every destination you might want to
connect to. Of course, it's pretty much impossible to build - there
aren't enough PCI slots for network cards in the back of the PC.

So once you accept that you can't build a full mesh, and will have to
accept less than a completely optimal and absolutly direct path, all it
then comes down to is how less optimal will you accept, and how much
traffic locality matters to you and you're willing to pay for.

Regards,
Mark.