Sunday 28 July 2013

NBN: Performance of Turnbull Fibre From The Node

There is a suggestion in the Turnbull Node Plan that "Fibre On Demand" is somehow identical to the current direct GPON connection of the current full FTTP Plan. This is not nearly the case, "Fibre isn't Fibre" when it's connected to a low-capacity Node.


Turnbull doesn't ever lay out just what his Node Plan does and does not provides:

  • Is the maximum distance, referred to as "a few hundred metres", 800m or more?
    • This sets both average/median access rates,
    • but how many ports are connected to each Node.
    • There is a world of difference in performance sharing a link between 150 and 400 ports.
  • How many Nodes will be eventually deployed by 2019 for "50Mbps to 90% of services"?
    • Is it 68,000 or 350,000?
    • How many customer ports per node will be supported?
  • What uplink bandwidth will Nodes be provided with?
    • Will that bandwidth be upgraded?
    • Under what conditions will links be upgraded? "Cost-Effective" is no answer.
    • What congestion levels will be deemed "commercially acceptable" by the Coalition?
      • Normally Nodes are connected by 1Gbps ethernet links.
      • While 10Gbps ethernet is available, Nodes that can handle that much dataflow are a great deal more expensive, by at least 5 times.
      • Will high-performance Nodes be installed everywhere or "cost-effectively", creating unavoidable network congestion for many subscribers.
If a Node has 200 ports installed and 150 active services, no matter what the ADSL/VDSL or GPON/Fibre access rate of the subscriber, they all have to share the uplink.

If the uplink is congested, it doesn't matter whether you have paid for 12Mbps ADSL or 100Mbps On-Demand Fibre, everyone will get the same low speed, poor throughput and lousy latency.

Those 150 active services can only get 6Mbps each sharing a 1Gbps uplink, notwithstanding having paid for a Fibre upgrade. The Node is a choke-point, it's designed that way.

We know from the long-running Gungahlin Experiment with ADSL over Telstra RIMs, that subscriber demand is highly correlated: there is a very strong morning and evening peak-period. It averages 2-4 times the low-load period.

This means that every wide-scale Broadband Customer Access Network (CAN) has to be dimensioned to cope with this peak-load. We know from the USA's Netflix (over 30% of night-time Internet traffic) that video streaming will severely increase this peak and its correlation: It'll be much worse when Netflix comes.

For the FTTN network to support the minimum almost-guaranteed access rate of 25Mbps, at least 4Gbps uplink would be needed in this typical Node.
But we don't know what Turnbull is going to buy: his guiding principle has been "Cost-Effective", a synonym for "Cheap". Because he's never addressed this issue, we can only assume it's Bad News: the cheapest, slowest backplanes and uplinks - 1Gbps.

The ACCC has had a lot to say about ISP's selling Broadband connections, especially their charges.
If NBN Co, via an Retailer, sells a subscriber a 100Mbps Fibre upgrade, they have a reasonable expectation that it will perform at that speed. Especially near that speed during the evening peak.

How many 100Mbps Fibre services can a single typical Node support? Just one or two.

But NBN Co are about to release 250/100, 500/200 and 1000/400 Mbps plans. At best, Nodes will only support 250Mbps, more likely they will be capped at 100/40 Mbps.

If that typical 150-port Node has 50 subscribers getting a full 50Mbps and averaging 30Mbps download during the evening peak, the uplink is already over-subscribed at 1.5Gbps. The other 100 subscribers only have to use 5Mbps, close to the current average download rate, for the uplink to be over-subscriber 2:1. That's a heavily congested link and unusable for everyone on the Node.

Right now, the average access rate is between 4-5Mbps, according to Akamai, so the maximum peak-hour demand per typical node would be at least 0.6Gbps if nothing changed. But the Turnbull Pseudo-Guarantee is for 25Mbps up to 50Mbps: at least 6-times the current rate, around 4 times the usual uplink speed of 1Gbps.

Is Turnbull planning to deliver Nodes that are 4-times too small on day 1. We have NO evidence to the contrary.

Adding a GPON Fibre connection will force a borderline uplink into meltdown. As I said earlier, bandwidth is shared, equally, and everyone suffers. This was the lesson from the Gungahlin Experiment.

It will be an expensive, slow process to upgrade 68,000, let alone 300,000 Nodes, from 1Gbps to 10Gbps, and even then will it be "too little, too late"? Not because "every link has 2 ends" and all the ethernet fibre transceivers (100-500,000!) will have to be replaced and upgraded, but because the Node "backplane", the electronics that shuffles all the data around, has to be replaced. This, like Vectoring, will cost a large fraction of the original Node cost.

The current GPON Fibre CAN is dimensioned to cope with very high peak-demand, with potentially congested links being able to be identified and upgraded, cheaply and easily. 

There will be no problem supplying individuals with 1Gbps services, nor a whole street with 100Mbps services.

That can't be said for the Turnbull Node Plan. Monitoring and managing Committed Information Rates (CIR) to each Node will consume a lot of resources, which the customer must pay for.

The only way to manage CIR demand is to limit access rates on Nodes. This translates to:
It's extremely unlikely that the Turnbull Nodes will ever support more than 100Mbps on Fibre and then only 2 or 3 services will be allowed per Node.
Fibre From The Node is a dupe: It's the most expensive solution possible, cannot perform the same as the current direct Fibre network and will be severely rationed.

5 comments:

  1. It is my understanding that the equivalent BT service in the UK will only provide 330Mb/s down - 30Mb/s up, thus negating any move to very high speed services available with GPON. This page http://www.thinkbroadband.com/news/5770-fibre-on-demand-openreach-confirm-29th-april-for-roll-out.html is instructive in explaining the associated costs - note that "In addition to a fixed installation fee of £500, a distance based construction charge will also apply, reflecting the costs of building a fibre network direct to a customer’s premise." Something Turnbull never talks about. FOD via FTTN nodes will ALWAYS be a second rate option.

    ReplyDelete
    Replies
    1. thanks very much, especially for URL that now sits with article :)

      Good to have independent, real-life corroboration.

      Delete
  2. On his blog MT says about 50,000 (178 premises / node)

    But still this is his personal blog not the policy...

    Mr Abbott says 10,000 - 20,000

    yeah i could say something but ill let people draw their own conclusions...

    ReplyDelete
  3. I'm still wondering about the physical logistics of how Nodes will be deployed. To meet the 25Mb min by 2016 would require X number of nodes (60k?) to be deployed EVERYWHERE in 2.5 years. To meet 50Mb min by 2019 would require possibly 3X that number of nodes. Now it won't be possible to deploy 3X nodes by 2016, so to me that would imply one of two options:

    1. After deploying nodes to meet 25Mb requirement, an entire revisit of the whole country would then take place to install those extra nodes to meet 50Mb; OR
    2. There will be some still to be invented commercially available "technological advance" that will allow the nodes placed at 25Mb spec distances to support 50+Mb minimum without needing to add extra nodes. That would still require a complete revisit to fit out the node cabinets with whatever that new tech happens to be.

    Malcolm has mentioned distances of 4-500m from node to premises, but I'm pretty sure that would be the 3X number for 50Mb speeds, so it can't be the initial 2016 build target. That would lead me to believe that Malcolms plan will involve two passes of the entire network to build out completely.

    ReplyDelete
  4. thinkanotherday, I think you have discovered the elephant in the room! I must correct you, though - you refer to "Malcolm's plan". Now I always think of a plan as something that has been thought through; is detailed; costed etc. Ready to go. Turnbull's "plan' is actually a "vague proposal" - a very different animal. And, as an increasing number of commentators are understanding, the LNP really have zero intention to ever turn that proposal into a plan. That's why the numbers in your post don't add up. They aren't intended to! All Turnbull thinks he has to do is to fool enough people that it is somehow safe to vote for the LNP and they will proceed with their "cheaper, faster" network. Once in power, of course, the ephemeral plan will disappear as fast as it appeared. The excuse will be that 'the coffers are bare', and as Steve says elsewhere in another blog, Telstra will be gifted the current NBN.

    ReplyDelete

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