Mobile stack needs overhaul
There is an inherent problem when the world of wired IP meets wireless. Wireless needs careful planning to keep loads within a comfort zone to prevent catastrophic failure when links get overloaded. On the other hand, TCP/IP by its nature saturates links by going faster and faster until there is an error before backing off. The entire paradigm from developers, carriers to telcos needs a redesign, but this will take generations.
Dr Neil Davies, founder and chief scientist of Predictable Network Solutions, explained to TelecomAsia that there was a crisis looming in networks which are in a death spiral of upgrades and investment while they have yet to recoup their costs to achieve return on investment.
“You cannot buy quality, like you cannot buy dark,” he said. Quality is an absence of problems.
The radical net neutralists want everything perfect for no money, but Dr Davies argues that quality is not equality of treatment, but equality of access to the same outcome.
The CDN is a good example that can create differential operational dynamics, not just savings. The UK regulator, Ofcom, came in and looked at what would be necessary to make CDNs network neutral, but it was too late. Their physical locations give them a direct anyone else [located elsewhere] as the delays are less. The only way to ensure net neutrality would be to add latency and make every competing CDN as bad as the worst.
Then there is a psychological aspect that needs to be considered. The key thing to get people using services is cognitive absorption. Davies shared a story of how one subject was talking over a video phone. Asked how long he was on the (very high quality) call, the subject said 45 minutes. In fact he had been on the call for 2.5 hours.
He had entered cognitive absorption, or in layman’s terms, he lost track of time.
Achieving cognitive absorption is well known to those involved in games. People play for longer. What they are measuring is the level of delay and loss after which the system starts to assert its existence into the minds of it users and stops people from getting into cognitive absorption.
Davies says that rather than design systems based on the idea of being plugged in to a wired network and spending enormous amounts of money later making it work on a mobile network, mobile apps should be designed with these loss and delay characteristics first, focusing on the attenuation of quality, the delta q, throughout the network.
Delta q conserves. It gets bigger the further you get away from the source as the loss goes up. If you get less delay, someone else has to get more. You can trade delta q but you cannot destroy it.
But here there is another problem. Davies said that there he has observed a sharp cut-off in the age of people who understand statistics. In the 70’s the royal society had a go at the government for lack of statistical teaching in primary schools in the UK, arguing that statistics and probability were key to being a scientist in the 21st century. The result is that many senior engineers end up crying trying to make their bosses understand.
And there is another crisis looming as today these topics have been taken out of the syllabus again as it was too difficult for kids to pass.
This manifests itself in the different mindset between TDM and IP. TDM was engineered to eliminate failures with a loss rate of 10^-15.
While for IP, the Ethernet engineer says, “Well, it works. I put a packet in and it comes out the other end, what more do you want?”
The internet was designed to be completely laissez faire, in the economic sense.
So what does this all mean?
For voice, a g711 codec works well if 98.5% of packets are delivered with between 60 to 70ms of overall delay variance. That means that it is possible to fit 101% capacity into a connection by randomly dropping 1% across the sessions.
Today telcos are just dumb pipes. Rather than focus on selling quality through these metrics, they have decided to build huge vertical stacks to compete with Skype. Rather than compete with Skype, it would be more simple to construct a system whereby Skype can work better and price for it. But just the type of app is not enough. A chat app for fire and rescue would need better quality than the same app being used by kids, for instance.
Davies advocates a three-class internet with different prices for each : Standard, superior and economy.
Backup or large downloads or new OS upgrades for phone can be put in economy class, for instance, and video or multiplayer gaming would go on the other end. The key difference is that rather than the app saying it has priority (which quickly leads to everyone claiming they have priority, which means nobody has priority) or analysis done by the network, it would be the app declaring what it wants and paying extra accordingly.
Today, in emergency situations, networks fail. But what they are actually doing is becoming fully saturated, still pushing as much data as they can, but losing enough data so that none of the apps actually work - which means that the data that was managed to be transmitted also is worthless and incomplete by the time it arrives.
Davies often measures ping times. Often, in mobile networks there is a disconnect and when it reconnects, sometimes minutes later, all the data queued up is re-sent. Only, the majority of the data sent is now worthless and yet it is hogging up valuable bandwidth that everyone is competing for. His worst ping time over a mobile network was enough for the message to travel to Jupiter and back.
In his Nivarna, Davies sees app developers buying quality and quantity at a published price to be part of the app offering itself.
But who is leading this change? Certainly not the telcos who are locked in a downward spiral of market acquisition with no price floor. They design for peak throughput and advertise that rather than see things as a data delivery market.
By publishing benchmarks, the regulator can help shape the market. These performance tables are used extensively by the telcos, even if it is for marketing, and they create their products accordingly.
But this change will be slow in coming. It is generational. it takes one generation to realise the situation, another to train the trainers and only then will enough people, be it in telcos or developers, will be aware of the need to reform the market that change will happen.