Wednesday, 21 August 2013

NBN: The Need for Speed I - the path to 1 gigabit services

I was asked to comment on a Turnbull question: "What will the average person/household ever need 1Gbps for?" There's a lot in there to unpack...

Right now, there are real uses for download speeds over 100Mbps and upload speeds of 100Mbps to match partners running 100Mbps down. Already Fibre is generating 85% more revenue per service than Copper, releasing higher speeds (for higher rentals) adds no costs and increases the revenue gap to 145% over copper. This is just good business.

While we're not very many years off Telepresence/Telemedicine applications needing real-time "haptic" (feel/force) sensors and stereoscopic HD video, that need the "trifecta" of high throughput, low-latency and zero-jitter to be rock-solid, reliable and usable professionally, there are hundreds of current uses for higher speeds.

Turnbull deliberately phrases questions, including this, to:
  • create false dichotomies
    • it's not 1Gbps but three new tiers, 250/100, 500/200, 1000/400, all at different price-points offering increasing value to customers that have the needs and means,
  • conflate terms,
    • especially meaningless undefined pseudo-scientific/statistical phrases like 'average' or 'Cost-Effective'
    • There is no "average" user. The demand distribution of broadband is highly skewed, an exponential, in fact.
      • Is 'the average' user the low 50% that use just 6.4% (six point four) of download?
      • Is it the people at 25 percentile that use the mean amount data?
      • Is it some fiction like "2 adults, 2 kids on average household earnings"?
  • confuse input and output
    • the government is running a business offering services to the public, not representing households to that business.
    • the governments first priority is to ensure that the business is fair to users, profitable and can in the long-term repay the whole of the government investment, not fail and cost the taxpayer tens of billions.
  • confuse wholesale and retail
    • the classic "$20,000/mth for 1Gbps"
    • he has implicitly included it here through his "$110 ARPU is a Great Big Price Hike" argument, when the reality is the exact opposite. It's achieved by giving consumers massive on-going discounts.
  • focus on irrelevant areas like 'households'
    • The chief beneficiaries and high-revenue (profitable) users will be business users, either small/home businesses or those for sometimes work from home. They calculate the costs and savings in time and pay accordingly. 
    • Most household are in the 75% who get a free-ride, getting NBN services at cost or below because the top 25% generate all the profit.
    • NBN is a business. It makes money by selling the same physical infrastructure at different rates, they offer real "model choice".
    • The top 25% high-demand users generate all their profits. Strategies to meet this demand and reduce "consumer surplus" are paramount.
  • create "strawman arguments"
    • it will be decades before 50% of premises use services that can only be delivered over 1Gbps.
    • well before then, the business the Liberals should be focussing on will be able to derive significant revenues from speeds over 100Mbps just by making them available.
  • hide and obfuscate the real issues:
    • customers derive real value from certainty and reliability: Fibre plans are for guaranteed speeds, both download and upload, it isn't "one price fits all" or "any speed you like so long as it's random".
    • the effect on NBN Co Revenue and
    • matching consumer demand with supply. The only economic way to supply high-speed services to the very few, very profitable users that will pay for higher speeds, is to rollout Fibre to everyone and then let them decide what they have the need and means for.
The most important misdirection of Turnbull is that the Fibre/Copper differences are about a lot more than the access rate headlined in the question:

  • 4-port NTD allows business/other sponsored connections for regular work or projects, free from interference with existing connections or forcing extra costs onto recipient.
  • Simple, quick, easy connection of additional services into unused NTD ports.
  • guaranteed quality of service through non-congenstion in NBN Customer Access Network
  • Access to higher-priority Traffic Classes, allowing real-time and streaming video to work reliably on all ISP networks
  • Guaranteed and selectable access rate, both upload and download speeds.
  • The focus on, and ability to deliver, rock-solid services of professional quality, suitable for all applications, versus the "everyone fights for packets on the same overstuffed link" of copper with only the default & lowest service class, TC4.

The whole point of tiered access is to provide real value, in the form of certainty and lower cost per Mbps, and model choice to consumer. In return, NBN Co increases its revenue from exactly the same physical infrastructure. It's just good business, which makes it very strange that the Coalition are fighting against it.

In April, NBN Co released the Revenue breakdown until 31-Mar-2013. At $29.62, it is already 85% higher than the probably $16/mth that will be charged for VDSL/Copper services.

If the 31% take-up for 100/40Mbps becomes 17% 100/40, 9% 250/100, 4% 500/200 and 1% 1000/400, the ARPU from access charges becomes $38.70, for no extra cost. This increases the Fibre advantage to 145% over he $16/mth for copper.

QoN 53: $29.62 AVC income

Business users put a price on their time and can then balance time savings against additional charges.
In the table below, I'll use $2/minute ($120/hr) as the minimum rate for business ($78,000 Avg earnings is about $60/hr, or $1/min. A business has to earn twice that to be profitable & sustainable).
I'll use 20 work-days per month.

I'll use a Retailer mark-up of 33% on Access charges. Because they bundle download volume and business users are looking for time savings, not high quotas, Retailers make significant money on unused download quotas.

The current AVC charge for 100/40Mbps is $38 wholesale, $50.50 retail (estimated). This will be used as the base for the additional cost.

To go from raw "Megabits to second" to a reasonable throughput (Megabytes/sec) value while allowing for packet overhead and some delay, I'll divide by 10, not the usual 8 bits per byte.
100/40Mbps is capable of 10MB/sec down and 4MB/sec up, or 0.6GB/min down and .24GB/min up.
This is the reference for the other plans.

Based on 30GB/month average download (ABS 8153 series, Dec 2012), the 30th percentile is the mean, or 30GB/mth. 50th percentile is 4GB/mth. Top 1% is 300GB/mth.
As of 31-mar-2013, NBN Co reports 31% take-up rate of 100/40.
No information on domestic/business split, nor on business demand and download/upload ratio.
There are 3+ million ABN's and an estimated 1 million small businesses around Australia.
As well, we know that there are a very large number of people who work an hour or more from home most nights

Break-even Savings Table at $2/min ($120/hr) 


Min/day needed-123.75
100/40 time-

Down GB/day-0.66GB1.32GB2.475GB
Down GB/mth-13.2GB26.4GB49.5GB
Percentile (est)-40%-ile32%-ile20%-ile

Up GB/day-0.264GB0.528GB0.99GB
Up GB/mth-5.28GB10.56GB20GB
Percentile (est)-47%-ile45%-ile35%-ile

60/40 Down/Up day -0.502GB1GB1.88GB
60/40 Down/Up mth-10GB20GB37.6GB
Percentile (est)-45%-ile35%-ile25%-ile

Conclusion: Right now, between one-half and one-third of all users, if businesses, would be save money buying for a link faster than 100/40. That's a very convincing business reason to make available the three speed tiers above 100/40Mbps.

There is a caveat: not all ISP's/Retailers with limited customers on a PoI will offer these higher speeds as soon as they become available. They may delay the availability per region, based on their "CVC" and backhaul provisioning in each location. As suppliers, they have an obligation to supply the service they advertise, which means having at least the CVC bandwidth of the maximum speed of their customers on each PoI. For the major carriers, this should not be an issue now.

Some applications, such as streaming a remote desktop, require higher upload speeds to match a partner's download speed: i.e. 250/100 is needed to stream to a 100/40 user.

Higher speeds are very useful right now in business and non-business video & stills production:

  • There is an increasing amount of video being shot and edited, by many individuals and organisations.
    • Hobbyists and volunteers devote considerable time and money to their work. They value their spare time very highly because they have such large time commitments elsewhere.
  • HD and 4K USB-3 cameras are cheap and common. These files, especially "raw format", get very large.
  • Avoids delivery of video project 'by petrol'. Decreases costs & time taken.

  • Raw video and images, especially photoshop with many layers are big: they need a lot of upload bandwidth when being sent
  • Important work, paid or not, cannot be replaced. People need at least 2, preferably 3 or 4, copies of it.
    • Flash drives failed, laptops & external drives are stolen, disasters/viruses happen,
  • nightly backups to 'the cloud' (dropbox or  'mega') of large files are really important. Reliably, fast uploads are critically important to an increasingly large number of people.

  • When discussing with clients questions on editing, they need to be able to see the screen in full-defnition and at full-framerate.
  • 'remote desktop' software can share whole screens, if the link is up to it.
  • This saves a lot of petrol and time, is useful right now, and gives better products.

Technical Background

1. There is NO average user

The download volume demand is highly skewed, exponential in fact. Top 1% consume 10% of download data, bottom 50% (half) consume 6.4% (six point four) - that's a 75-fold difference between the aggregate use of the two groups, 300-1000 fold between the top 1% and bottom 1%.

Mean is at around the 25-30%percentile. Which fits the 31% take-up rate of 100/40 Mbps of April 2013.

2. Download volumes don't necessarily translate to demand for raw access rate.

If your work is browsing and upload/download of work units, waiting costs you time which equals dollars for someone. At $60-$150/hr ($1-2.50/min), you don't have to save much time every day to pay for a high speed link.

 If you 'just' stream video or download know files, you can get away with a comparatively small pipe

 If you have a server that backs-up overnight, it needs enough upload bandwidth to finish in its 'time window'.

  • Raw line access rate isn't what the ISP/Retailer can actually deliver through the whole chain.
  • This isn't sustained throughput & latency (or round-trip time/delay)
  • ISP's make their money through congestion. They have 'contention' ratios of 60-100:1.
    • 'contention' also called 'over-subscription', the ratio of backhaul to sum of all customer link speeds]

3. Only 0.1% of people 'need' more than 1Mbps for an average 30Gb/mth download. [false meme]
The average for the low half of users is 384GB or 14.6kbps continuous (24/7). Full developed in a later piece.

This addresses a fundamental problem of Telecommunications:
all individual's traffic is very bursty & highly variable in demand.This is more so with Broadband.
Addressing the Need for Speed, Answers:

1. Throughput (download volume)

 Right now, home businesses and people who take work home, can cost-justify higher rates, as per the "break-even table".

Domestic streaming of 4K TV will also benefit from 100/40 and 1000/40. More "headroom" on downlink and uplink gives a better service, all other things being equal.

There's a subtle problem for ADSL/VDSL using current encapsulated "PPPoE" (versus a direct IP connection to the Internet on Fibre, packets are hidden until in the ISP premises). The download speed is controlled by the available upload bandwidth, this is worse with larger asymmetry between Download and Upload speeds. Each packet has to be "acknowledged" when received, so the sender doesn't flood the link or run too slowly. If the uplink is 100% busy, especially with Big packets (at 1Mbps, 1500by has packetisation delay of 15msec, an age in network terms).

If someone decides to upload 100MB of pictures or video during peak TV time, the download speed of a 12/1Mbps service effectively become 1-4Mbps with at least 8Mbps inaccessible: little 'ack' packets have to wait in line behind the monster upload packets.

2. Latency and Round-trip time (delay) - relied on by Real-Time services (video, telephone, ...)

The flipside of Throughput is 'latency', either one-way or round-trip. The time it takes packets to reach their destination and for acknowledgements, if sent, to return.
But high access rate does not translate to low latency (you want low latency, just talk to someone on a Satellite link. Speed is OK, latency travelling to stationary orbit and back is around a second.)

This is why NBN Co offers four Traffic Classes (TC1-TC4), or priorities.
These are 2-bits in the packet header, the routers first send all TC1, then all TC2, ... and lastly all TC4 packets that it has stored in its buffers, waiting to 'get on' the backhaul link to the ISP.

Normal data is the default "TC4" - lowest priority.

High-speed links don't correlate that well to low-latency, but there is one effect linking speed and latency: packetisation delay.

 One component of latency is "packetisation delay". It affects every single 'hop'. Packets are sent, stored and checked by the next device in the chain, then sent onto the next device until after 15 or more hops, they arrive on the other side of the world. But the speed of intermediate links can be very fast: 100Gbps.

 Because there are many hops and each hop, roughly, equates to two full packetisation delays - the time it takes to put the full packet, data + headers, "on the wire". At 12/1Mbps a 1500byte (1.5kB) packet will take 1.25milliseconds just to send down the wire. It's 0.156msec for 100/40Mbps and 0.0156msec for 1000/400Mbps.

 There are many things that affect latency, mostly under the control of the ISP, some are 'upstream' and out of their control (eg far-end and international links). Which is where the traffic prioritisation comes in... You can pay for priority for your packets if you have the need and means. This isn't available in the design of VDSL2 as released.

 There are many applications that need low and consistent latency ( or low "jitter"): voice (telephony) and streaming video/media, especially multicast. Which is why the embedded Telephone Adapters in Fibre NTD's run at TC1 and multicast video is broadcast at TC2.

The slam-dunk for Fibre over Copper is the traffic prioritisation. The current VDSL model will give everyone a single link at TC4.

New applications like interactive maps, Geographical Information Systems, Robotics (telemedicine and more) and 'telepresence' require guaranteed low-latency or things "turn to mush" very quickly. You don't want to be in the middle of sensitive operation, surgery or flying a remote aircraft and have "buffering" messages appear.

Telemedicine, if it includes HD-video communications (monaural or stereo), needs both low-latency and high-bandwidth.

The 5%-10% (?) of patients with chronic illnesses already consume 50% of GP services and 70% of total health expenditure. Being able to better monitor and respond quickly to conditions of concern is vital to lowering our $120+ billion health care budget

The Department of Health can spend $30/mth on 25/5Mbps service for pensioners with many GB included and supply & install a router, laptop or tablet, TV+camera and support them for $500-$750/year. That's around one tenth of a day in Hospital.

With a 4-port Fibre NTD, this is trivial for a semi-skilled worker, like a community nurse, to install and get right, first time. There is no problem with link speed or VDSL ISP or errors/dropouts, usage caps or the person (on a pension) not being able to afford it...

No comments:

Post a Comment

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