[AusNOG] [ISOC-AU-mems] Quigley announces architectural "stake in the ground"

[Mark Smith]

Paul Brooks wrote:
> Adrian Chadd wrote:
>> On Wed, Sep 16, 2009, Mark Smith wrote:
>>
>>   
>>> The dumber the better ... ideally support for direct customer site to 
>>> wholesale aggregation site larger than 1500 byte Ethernet frames.
>>> (e.g. >1500 to allow for options of MPLS, L2TPv3, Q-in-Q, Mac-in-MAC, 
>>> PBB, PBT etc. encapsulation, just in case)
>>>     
>>
>> So what about direct customer<->customer communication? Does that have
>> to pass through your aggregation layer(s) ?
>>   
> 
> If nothing else, the lawful intercept obligations would suggest yes. I 
> can't see these requirements being relaxed in the NBN world.
> 

The more interesting question is who's is the LI obligation going to be? 
The NBN or the customer of the NBN? It'll be easier if it is the NBN, 
however that then may start placing technical constraints on innovation 
opportunities for ISPs. For example, if the NBN mandated L2TPv2 because 
that's how they've decided to perform the LI function, then if we wanted 
to do anything else other than PPP layer 2 encapsulation, we have to 
start doing things like e.g. MPLS over GRE over IPv4 over PPP. That's a 
big shoehorn to allow us to run MPLS between us and our customers (SMEs 
e.g.). They might also have to prohibit us from doing that too, as their 
LI devices might not understand that many or those specific encapsulations.

> This need not be a significant performance penalty - if the access 
> network is some form of PON, this looks much like a DSLAM with virtual 
> point-to-point links to each home in any case. The upstream packets from 
> one home have to go up to the active OLT before they can be turned 
> around and sent back to another home - the backplane of the OLT, or a 
> big ethernet switch nearby can be treated as an aggregation point with 
> negligable time penalty.

Agree. Typically when people talk about aggregation in this LI context, 
they're referring to layer 3 aggregation, which then means mandatory and 
inefficient hair pinning of traffic back to the layer 3 device, even 
though the nodes may be layer 2 adjacent, and could communicate with 
each other directly, if they are made members of the same subnet. I 
don't think the fundamental LI functional requirement mandates that 
hair-pinned layer 3 architecture, and ideally LI should be performed as 
close to the customer as possible and at layer 2 i.e. at the customer 
specific service delivery point (on their ADSL circuit for example).


(Personal opinion, not speaking for Adam)