[AusNOG] ubiquitous peering

[Matthew Moyle-Croft]

On 18/12/2012, at 10:38 PM, Jake Anderson <yahoo at vapourforge.com> wrote:

> So I was wondering and the list seems quiet.
> With P2P content (games, skype etc not just torrents) soaking a decent amount of traffic, and the NBN having relatively few "exchanges" if you will, I wonder about the possibility of peering at that level.
> 
> IE within each POI everybody sees if the traffic actually needs to leave said POI.
> Presumably routers and such would need to be configured automatically but I wonder how much of a gain there would be from it.
> 

It's all about the relative costs.

Of the 121 POIs a large number (80 or so IIRC) are going to be in metro locations - so the cost of dark fibre is going to be in the few thousand a month price bracket to aggregate it back to a central location where the bandwidth exchanged already justifies an IX.

The latency gain and cost benefit of pushing your L3 devices to the edge reduces rapidly as the cost to aggregate back into a central location in a metro drops as well as your relative market size in a metro.

(Think about the numbers when a POI has 100k services, a service might only use 300GB/month maximum, that you don't have 100% of the market and the cost of fibre/backhaul drops but the interconnect rate with NBN doesn't - the numbers of mbps you might share drops quickly and makes the maths hard).

Some of the more remote POIs (eg. Darwin) might gain from this - but typically I can see metros like that gaining a meeting location for an IX anyway over time.  

Don't get me wrong - in a lot of places I think this'd be good to do - but reality is that it needs to be more than a good idea, it usually has to have some rational business sense behind it to get people to buy into it.  I don't see that.  (I've done the maths on this a number of times and can't quite make it work for most locations).

MMC