Thursday, 18 July 2013

NBN: Copper or Fibre? Progress of speeds.

The point about an Optical Fibre Loop, or Fibre Customer Access Network (CAN), is that it has near-infinite growth potential for very low incremental costs. Moore's Law is driving transceiver speeds and large-scale commodity production is driving per-part pricing down. Another Silicon Technology Perfect Storm to be ridden, for some, to undreamed wealth, and for those stuck in the past, to be smashed.

The commercial question is:
How can Network Owners exploit this capacity/cost relationship and turn it into increasing real-revenues for the next few decades?
Proponents of Copper CANs squeal about achieving 100Mbps, Real Soon Now, like it's noteworthy or Really Useful. Then they excitedly babble over the announcement of laboratory experiments supporting a whole gigabit on twisted-pair or "G5 Radio". Apparently cost and availability are not commercial realities to them.

Here are some solid, commercial reality, not a pipe-dream.
Right now, CISCO will sell you DWDM 100Gbps cards that will give you 9.6Tbps over a single fibre for thousands of kilometres:
The new CP-DQPSK modulation supports 9.6 Tbps capacity transmission over Ultra-Long-Haul (ULH) networks of up to 3000 km of unregenerated optical spans.
 They won't be cheap, even for Telcos. But they're here, they are in production and they work.

How to compare the evolution/progress of the two technologies?
Ethernet, the commodity data networking in computing, offers a reasonable basis.

The rough history of Optical Fibre in Ethernet is [source]:

  • pre-1993, 10Mbps Fibre Optic Transceivers as device-link extenders.
  • 1993, 10BASE-F, 10Mbps
  • 1995, 100BASE-FX, 100Mbps
  • 1999, 1000BASE-X, 1Gbps
  • 2003, 10GBASE-SR/LR/ER/ZX, 10Gbps up to 40km
  • 2010, 40GBASE-x and 100GBASE-x, 40Gbps and 100Gbps. at 10 & 40km.

Wave Division Multiplexing, Coarse (16) and Dense (32-96), came of age in the last decade.
It sits outside the 802.3 specifications, allowing Data Networks to achieve massive throughput.

Twisted-Pair, remembering this is NOT the Telecomms grade 0.40mm or 0.60mm single-pair, but much more expensive Cat 4/5/6, using 4-pairs:

1990, 10BASE-T, 10Mbps, 100m
1995, 100BASE-T, 100Mbps, 100m (reports of 1000m @ 100Mbps existed in 2005)
1999, 1000BASE-T, 1Gbps, 100m [the default chips in PC's and laptops: 10/100/1000]
2006, 10GBASE-T, 10Gbps 55-100m [very rare. More common is SFP+ over 1-10m]
2010, 100 Gbps, 1-10m. Printed Circuit boards and Infiniband CX-4 cable. not twisted pair.

Twisted-pair did very well for a decade, going from 10Mbps on Cat-3 cable, to 1Gbps on Cat-5a/6 cable. It is still the backbone of commodity computing and low-end servers.

Copper improved in speed by a factor of 100 in just 10 years: let's be generous and call that 7-doublings (128 times). That's 1.4 years per doubling, which is pretty impressive. Then it stopped.

By these standards, we'd expect to 100-250Gbps copper ethernet defined and in-use now.
While a 10Gbps copper ethernet standard appeared, not in 2001/2, but 5 years late, it has no market-penetration. Anyone who's serious are using either Fibre (SFP) or SFP+ cables (same socket as a Fibre transceiver).

The first significant Ethernet over Fibre standard was 1Gbps in 1999.
Take the CISCO card above, with 9.6Tbps max, as the 2013 reference, and we have a 9600-fold increase in speed: or over 13-doublings.

For more than a decade, production Optical Fibre has doubled in capacity every year.
Yes, like CPU's, it now relies more on parallel channels than raw link-speed, but it does deliver.

This isn't some theoretical laboratory demonstration that might, just might, appear in consumer products in 10-15 years time (there are many more failed ideas than real products).

This is for-real, available products in production, not vapour-ware or GeeWhiz! hype and speculation.

The rule in I.T. for many decades is simple: What we see at in the high-end today, we'll see in volume production for commodity devices in ~5 years. Prices will come down over time.

Before 2020, the notional end of NBN Co's construction project, there will be available 100Gbps commodity transceivers for residential use. High-end consumers, business or domestic, will be able to upgrade to DWDM, to purchase Terrabits of bandwidth.

The point about all that speed: it will cost what today's 1Gbps electronics cost.
The per-bit/second cost in the CAN by 2020 will have dropped 100 or more times.
Not theoretically, but for-real.

In the meantime, Mr Broadband, Malcolm Turnbull, assures us that he'll absolutely be able to deliver 100Mbps over really bad copper to some, not all, of the people that want it, for an extra-ordinarily high price - on equipment that is yet to support large-scale networks, from vendors who see the end of the road and will finely judge just how much they can squeeze their customers.
These fancy copper interfaces are everything that fibre transceivers are not: Not cheap, not commodity, not volume production, not a competitive market.

Right now, we know that nearly 95% of network data traffic is generated by half the customers, and 1% of customers generate 10% of all volume.

These are the profitable customers, they will drive the network growth and drive down speed and volume charges for the majority. With xDSL/FTTN, their demand cannot be matched with supplied speeds - their business will be lost, permanently.

What sane business model deliberately drives away the most profitable customers like this?

It's trivial, fast and cheap, with an Optical Fibre CAN, to keep upselling these high-end consumers for decades. We know this from the evolution of PC's across 3 decades. These few customers are the major economic and financial driver of the new Data Communications Networks. Serving them is the high-profit, high-volume side of the business. The rest is a low-profit, commercial distraction.

This is a problem of Good Business and Intelligent Application of capital, not of trenchant ideology.

It's not a question of "How will anyone ever use that much bandwidth?" or "What will the next Killer Application be (to drive demand)?".

The leading-edge users, have, and will always, discover ways to use more bandwidth, more storage and more CPU-cycles.

That's the big lesson from 65 years of Computing/I.T. history. It's been re-proven many times over:
Demand always exceeds Supply. If you install The Behemoth designed to serve all your needs for a decade, it will be stretched past capacity in under a year.
The Network Operators don't have to guess "What will drive demand?", they only have to be able to supply increased demand, anywhere it is sought, at a small incremental cost.

That's a precise match for Optical Fibre CAN and a complete FAIL for the existing Copper CAN.


  1. All excellent points. But, mostly not relevant.

    It is worth noting that the NBN Co is actually deploying GPON. The GPON standard (G.984 and their Alcatel sourced equipment) was developed in the late 1990's, and hasn't been improved upon since then.

    Because of budgetary constraints and equipment standardisation issues we can expect that NBN Co will STILL be rolling out GPON in 2021. There is no option for a technology refresh in their business plan.

    There are many ways to skin the cat. NBN Co has chosen a very blunt rusty knife to do the job.

    Forward thinking network operators are deploying direct fibre, and active fibre technologies that can be upgraded easily over the lifetime of the fibre.

    1. Yes, I've read WIkipedia, but thanks for trying to educate.

      While NBN may not come back to upgrade NTD's, they have written about upgrading in-place.

      My points was that GPON @ 2.5Mbps downstream will be upgraded, substantially.
      And the chips at the heart of the system the lasers and photo-diodes, will be commodity. They'll be borrowed and used in later versions of GPON - just like 1000BASE-FX borrowed the physical layer of FCAL. (Fibre Channel).

      I won't debate active ethernet vs GPON. Both have their place, both have strengths and weaknesses.Could the splitters in GPON be upgraded to DWDM and we have pure ethernet from the FSAM to premise? I don't know enough of the technology. But you end up with what looks like a Node - an active device in the wild outdoors. Even street lights have problems being exposed to the elements...

  2. Just pointing out that in contrast to your argument about the development of optical transmission speeds (which is 100% correct), GPON has NOT been upgraded since 1990s. It is a "dead" technology, stuck in the 1990s.

    A GPON rollout cannot be upgraded without replacing all of the installed NTDs, along with inherent service interruptions. The existing GPON NTDs are not colour aware, so dense (or coarse) WDM can't be simply overlaid, and fitting into the timeslots is also fraught with difficulty. If it were simple, it would have been done and it would have been standardised 10 years ago.

    Check the developments in DOCSIS performance if you want to see how an upgradeable system, Hybrid Fibre Coax, has been accelerating over the past 20 years. It is more in line with your generic description.

    Also, NBN Co don't have the luxury to go back and update their completed builds to new technologies. The "ubiquity" principal of their wholesale service provision ensures that we'll all be getting the same old GPON through to the end of their build.

    So, writing about the hyperbolic rate of development in optical transmission, and equating it to the NBN Co rollout, is missing the real point, imho.

    The real point is that a lot of money is being spent on building something that is full of vendor and technology lock-in, and could easily be done so much better.

    Have a look at Simon Hackett's recent video on the issues.
    It is worth 20 minutes of your time.

    1. I disagree with your assertion "could easily be done so much better" - the essence is doing it Professionally. That means it's robust, reliable, scaleable and upgradeable, will last 50 years AND survive being worked on by sometimes disengaged and unfocussed linesmen, especially if they're hung-over.

      Easy to build something quickly, harder to build something that will "Just Work" and work effortlessly for decades on end.

      Yes, Simon Hackett's presentation is good. I disagree with his analysis of the $110 ARPU.
      That's mostly carried by the top 10% of users. Everyone else is getting _great_ discounts.

      The real value in Simon's presentation is that someone credible has come out and resaid what Quigley said earlier this year: we really need to have a non-partisan debate on what we need and how we can get there...

      Simon has some good ideas, but mostly along the lines of "do more of the same".
      I think a 4-port VLAN NTD is a killer App for many users. But why make it the default and give it away for free.
      Allow RSP's to supply their own, fully-managed, GPON router. Yep, absolutely.
      But don't make that the One and Only option...


Note: only a member of this blog may post a comment.