Internode Blog

NBN Points of Interconnect and the future of competition

Sunday, December 19th, 2010 by

(Edited to add an epilogue on 20th December 2010)

A key sticking point in the context of the National Broadband Network is the question of how many exit points the network should have.

This post provides my point of view about that question, and argues that competition (and consumer benefit) is critically dependent on the number of POI’s being minimised.

I’ll need to define some NBNCo (and other) acronyms for the following to make sense, so here goes:

  • RSP: Retail Service Provider – one of the companies (like Internode) who are providing retail telecommunications services using the NBN as one of their wholesale network paths to connect to the homes or offices of their customers.
  • POI: (NBNCo pronounces like the “Koi” – yes, the fish) is a Point Of Interconnect – a physical location where an RSP links their existing network to the emerging NBN network. A meeting place between the NBN and the people who complete the circuits concerned, end to end, to create something functional that a customer can buy.
  • ATM: Asynchronous Transfer Mode – A protocol for moving data over network links that embodied the interesting innovation of using a very large number of very small data packets, called ‘cells’.
  • ACCC: Australian Competition and Consumer Commission – If the Trade Practices Act was software, the ACCC would be the CPU than runs it. Their role is to implement (and where needed, prosecute) the Trade Practices Act.

The debate over picking the ‘right’ number of POI’s for the NBN has some significant parallels to the manner in which the size of ATM ‘cells’ was chosen. Specifically, despite strong technical arguments, the ultimate choice was a politically motivated compromise that satisfied nobody, but which also minimised offence to anyone at the time the choice was made.

Politically motivated decisions on technical and access equity questions are generally a bad idea.

Here is the short version (from of how the ATM standard came to have a compromise cell size:

ATM broke up all packets, data, and voice streams into 48-byte chunks, adding a 5-byte routing header to each one so that they could be reassembled later. The choice of 48 bytes was political rather than technical. When the CCITT was standardizing ATM, parties from the United States wanted a 64-byte payload because this was felt to be a good compromise in larger payloads optimized for data transmission and shorter payloads optimized for real-time applications like voice; parties from Europe wanted 32-byte payloads because the small size (and therefore short transmission times) simplify voice applications with respect to echo cancellation. Most of the European parties eventually came around to the arguments made by the Americans, but France and a few others held out for a shorter cell length. With 32 bytes, France would have been able to implement an ATM-based voice network with calls from one end of France to the other requiring no echo cancellation. 48 bytes (plus 5 header bytes = 53) was chosen as a compromise between the two sides.

In the NBN, following a public consultation process, the ultimate choice of POI location and number is apparently being discussed behind closed doors right now by the government (presumably in conversation with moral equivalent here of France in the ATM story – Telstra) at this time.

The clear choice of NBNCo (and smaller RSPs) is for there to be a small number of POI’s (circa 14) located in capital cities. A small number of POI’s in capital city locations means dramatically lower economic and technical barriers to entry and participation for all RSPs. It will lead to a larger number of more diverse service providers (and services) on the network.

The opposite view is held by the very largest RSP’s (Telstra and Optus especially), who have been lobbying for there to be a much larger number of POI’s (circa 200). This is not surprising, as these are the players who own their own extensive fibre backbone networks that will let them plug into 200 POI’s without significant investment.

The actual location of these 200-ish POI’s (a list I’ve not seen, though I’d love to see it) are presumably ones that conveniently coincide with the existing Telstra and Optus networks, meaning they will provide essentially zero cost interfacing to the NBN for those larger players.

The ‘many POI’ model then means that all smaller players will be forced to buy access from their own (generally) capital city based networks, through to each of those 200 POIs, from one of the few players with that existing fibre backbone structure in place.

The self-serving benefit of this for those larger players is obvious.

They will also permanently have their fingers in the value chain of everyone else in the industry (something the NBN was intended to avoid) by charging all the other RSPs to reach the NBN POI’s via their fibre backbones. Because there will be very few of the big players, the access pricing to access the NBN POI’s will tend toward cartel behaviour (high price with no economic driver toward reducing it over time).

The last decade has provided a signature example of this behaviour in the costs of a Telstra Wholesale construct (of a very similar sort) called an “AGVC”. Access costs for this existing economic access ‘moat’ are a major cost input in retail ADSL2+ services, and these costs have essentially not lowered in real terms in the last decade, while all other end to end network access costs have dramatically improved over that period. In effect, Telstra uses its wholesale division to transfer profits from the rest of the industry into itself.

The last decade has also shown that where cartel-like behaviour occurs in the telecommunications industry, it can be extremely long lived. The signature example here is the ‘Gang of Four’ Peering cartel that has existed in Australia for most of the last decade (though in practice it is pretty much down to the ‘Gang of Two’ now – Telstra and Optus).

The ACCC accidentally created this cartel while trying to enhance peering between ISP points of interconnect a long time ago (its a long story). Since then, the ACCC have been frustratingly unprepared to fix their own mistake, and have allowed Telstra and Optus to charge the rest of the industry for network interconnect access between the Telstra and Optus networks and other ISP backbones to this day.

(This also leads to an obvious requirement for the preservation of competitive pricing and access efficiency, which is that all RSPs should be contractually required to neutrally peer their IP backbones as a condition of being able to access the NBN).

The commercial interests of Telstra and Optus appear, disturbingly, to be more important to the government than keeping the NBN playing field level.

And now we can return to the earlier example of politically motivated compromise in the design of ATM, and where failing to learn from the mistakes of the past may impinge upon the future of the National Broadband Network.

The media has reported lately that there is a likely compromise announcement on the way from the ACCC and the government, in which the failure to decide between the 14 POI model and the 200 POI model may be resolved, incredibly, by mandating some number of POI’s in the middle, between those two extremes.

That would represent a decision that can only be framed (like the ATM cell size) as being politically motivated – and the classic mistake in politics of choosing the average between two opposing views instead of being brave enough to make an appropriate decision.

And like the ATM cell size debacle, the outcome would be bad for everyone. It would suit neither opposing POI camp.

It  would ‘half fix’ the commercial drivers of the big guys, while ‘half hurting’ everyone else. Permanently. And in a way that may be impossible to fix later on.

Not surprisingly, I’m strongly in favour of the 14 POI’s model. Its quite insane in my view for the NBNCo to spend $43bn (or whatever it does cost) on the network, to bring it 99% of the way to where all the RSPs are generally located (in capital cities), and to leave that last piece of fibre backhaul expenditure out there to have to be contended for in parallel by every RSP.

The NBN is a massive market intervention. There is no escaping that. It will create winners and losers. There is no escaping that either.

But if the NBN is to truly be the new network offering wholesale open access nationally, then the Long Term Interests of Endusers are best served by having more choices of providers, and by having the POI’s for the NBN structured in a manner that does not allow smaller competitors to be the permanent economic victims of larger ones.


Epilogue: 20th December 2010

Today (20th December 2010) the NBNCo released its 3 year corporate plan and proposed wholesale product and pricing overview to the public.

It includes a decision that looks to have been driven by the ACCC and/or the government to have… 120 POI’s…  half way between 14 and 200.

Obviously this is not my preferred outcome.

In particular, 120 points of interconnect will generate higher setup and running costs for all industry players (except Telstra, of course, as it looks like these points will all be Telstra exchanges, in addition to Telstra being paid off to shut down its copper network).

At a deep level it seems (to my way of thinking) as if the 120 POIs model is a bit of an artificial construct, keeping an existing backhaul market supported (to an extent) in the new world order.

In practice, this will raise end user retail costs for competitors to Telstra, because the additional costs (both fixed and recurrent) to build, operate, and maintain 120 points of interconnect will have to be bourne by all industry players (except Telstra) in addition to the NBNCo wholesale charges, Internet supply, national and international network link costs, and of course support/staff costs and other business overheads.

The NBNCo wholesale access costs also appear to be for a PSTN+Data bundle, so in that context the cost inputs will need to be evaluated against the bundled cost of voice + Internet access in offerings in the market today.

Remember too that the NBNCo pricing is a wholesale input cost (ex GST), but retail prices are ‘including GST’ numbers.

Its also worth appreciating that the cost estimates in the NBNCo business case (of the per-user costs) are based on steady-state customer base sizes and it will take ten years, and a lot of money for each ISP to get ‘there’ from ‘here’ across that ramp-up period.

The effective per-user cost during that decade of ramp-up will be much higher, as each national RSP will have to buy 120 CVC’s with an initial  200 megabits of capacity each at $20 per megabit, and then slowly fill those CVC’s with paying customers.

The reason why the minimum viable CVC size is 200 Megabits is this:

To service 100M link speed customers, you need at least a 100M ‘burst’ capacity above the 100M line rate to avoid immediate issues with contention in peak periods – even with a small number of customers. Once there are enough customers to properly ‘fill’ 200M in a CVC on an average basis, the amount of burst capacity one has to purchase will begin to diminish as a relative proportion of the total. However, doing the sums (dividing customer base numbers nationally across 120 CVC’s), this will only happen properly for an ISP with hundreds of thousands of customers nationally.

(We learnt this lesson the hard way, through our work in NBN Tas. 100M of CVC isn’t enough, even with a relatively small number of customers, when those customers have 100M port speeds).

Unless I’m missing something, this suggests the initial national overhead for CVC fees may be 120 (CVCs) x 200 Mbits x $20/Megabit = $480,000 per month + GST (and that is just for the CVC’s – its not to install, operate, and pay for backhaul out of the 120 POI’s concerned, an additional amount which is probably of a similar magnitude).

(There are some worked examples of national average per-user running cost for various sizes of customer base in the NBNCo documents released today, but as noted above, they are ‘steady state’, end-point average per-user costs, at the end of the ramp up period, on a ‘full’ network. They don’t show how these per-user charges will start far higher on day 1 and ultimately converge on the numbers in those worked examples only as a terminal state)

This large number of points of interconnect will, therefore, drive the industry toward having a smaller number of bigger players, rather than a larger number of smaller players, as the overheads of operating (and ramping up to survive all the way to that terminal size) in 120 POI’s may be too much for very small players to afford.

Internode is large enough to do this, but I am a bit concerned for anyone who is much smaller than we are.

As consumers, we will also have to hope that this doesn’t then end up with ‘super-consolidation’, much like the banking industry.

As a general statement, though, its great to see the pricing out in the wild, and the industry will undoubtedly start to do some proper modelling of expected operating costs based on these prices over the coming months. NBNCo have been highly consultative, and so I’m quite sure they’ll listen to views on their pricing with interest (now that it is released).

Remember also that these prices don’t yet apply to the existing Tasmanian NBN (which is running on pre-commercial pricing until half way through 2011).

In summary, then, the industry has a little while to get its head around the pricing chain that it now has in its hands, and all thoughts on it at this point are necessarily preliminary ones.

At this point, the industry can work on developing reasonable retail cost models toward the expected deployment of mainland NBN sites in mid to late 2011.

However you slice it, a wholesale open access network with transparency on pricing is a massive step forward compared to an industry currently dominated by the ghost of Telecom Australia.

Post tags: , , , ,