Vodafone publishes high-fidelity network measurements using ∆Q metrics

Vodafone has published network data using state-of-the-art high-fidelity measurement and ∆Q metrics. I highlight some of the key slides.

The group head of fixed access for Vodafone, Gavin Young, gave a presentation last week to the UK Spectrum Policy Forum. The full set of presentations from the session are available online on the techUK website, and you can view Gavin’s directly on SlideShare.

Below are some of the key slides with my personal commentary added. I have previously published an interview with Gavin on Demand Attentive Networks, and you may wish to consider these slides in the context of that interview and the associated IET report.

This is the first time a major operator has published ∆Q-based data, but Vodafone is not the first operator to have engaged in high-fidelity measurement. For instance, there is an earlier BT case study. As for the many others, blessed are the NDAs, for they shall be respected.

As I and colleagues presented to Ofcom five years ago, the broadband market is going through a series of stages of evolution. We might debate their exact names and demarcation, and they are very real nonetheless. Quality is the next one, as it is a prerequisite for delivering assured cloud application access. This has more value than undifferentiated and unassured generic broadband Internet access.

You will note that unusual keyword in small print in the middle: stationarity. I told you so: stationarity is the new speed. I hope equipment vendors reading this have your stationarity tests lined up, so you can still sell your network equipment to tier 1 operators in future. It can’t be long until wise regulators are testing broadband services for (non-)stationarity too.

This term is common knowledge in the financial services industry, since the value of predictive investment models depends on the underlying statistical stability of any presumed cause and effect. As statistically-multiplexed networks move from a taxi-like ‘pipe’ to an Uber-like ‘resource trading’ model, only those who understand this basic concept will prosper.

The “killer app” is humans engaged in all forms of sensual social intercourse. Voice communication is one, and there are many others. Telephony is a virtual reality application, and “tactile” comms technology is 150 years old. After 100+ years of engineering success, the Internet lost the performance isolation between data flows that the telephone system had. “Best effort” (i.e. you get what you get) has remained poor at tactile applications, and always will be.

The core proposition of “unified” communications has always struck me as a conceit. “Unification” is a solution to a vendor problem more than a user one. It’s like putting the toilet, shower, kitchen sink, dishwasher, washing machine, bath, dog’s drinking bowl and garden hosepipe in one room to “unify” the water-related activities. So I don’t see UC as being more than an interim technology solution to the human co-presence problem.

Most communications is dyadic (2 people) and better forms of virtual and augmented reality will be the true replacement or upgrade for telephony when it becomes end-of-life. Amazon Echo is your ambient AR phone, and Magic Leap may well be your next wearable one. These need a new low-latency network with the right quality properties. (Hint: has to be sufficiently stationary. Got it yet?)

Meeting these far stricter performance needs means that the marvel of today’s packet networks is not enough. Users are demanding the miracle of it “just like being there” (if not even better), and we’ve got to find a way to supply that if we wish to satisfy them. The hypervoice VR conference call of the future is a custom-simulated space with nice pictures and gentle music (unique to you to get the mood right), filled with avatars of your colleagues, all in a “show and tell” mode as you discuss some virtual object or process.

Quality does not come from quantity. That is not news to any longtime reader. You can’t make an apple pie from rotten apples, no matter how many you have.

“Quality attenuation” is the packet loss and delay when seen as a single unified concept, and is quantified using probabilistic ∆Q metrics. The three basic components of ∆Q are (G)eography, packet (S)erialisation [and deserialisation], and (V)ariable contention delay. These give different bearers very different quality characteristics, even if they have the same nominal bandwidth.

Here you can see the breakdown of ∆Q into the G (green), S (blue) and V (red mean and yellow stddev). Note how the G dominates the S, and making links faster has rapidly diminishing returns in the overall latency scheme of things. However, the real story is that the experience is now dominated by the (tail of) V, which is much bigger than G or S.

G isn’t changing much, as the speed of light remains stubbornly fixed. Whilst most of the telecoms industry is sinking pointless capital into capacity to reduce an already trivial S, some of us are instead working to schedule that V to dump it onto the flows that can most tolerate the hit. This is the essence of the little-noticed fundamental shift of telecoms into its third epoch.

Finally, you can see a comparison between different bearer technologies, namely VDSL (dedicated copper loop) and GPON (shared fiber access). Whilst the high-speed fiber has a smaller S (in blue), that’s a pretty trivial part of its value.

In this example, moving the content or computation closer to lower the green G has had a big effect, hence the “edge compute” trend. The V is also a lot lower on GPON here (possibly as the capacity is largely idle), and is delivering better stationarity. As a result, this bearer technology is likely to be excellent for real-time and tactile interactive applications.

I spoke to Gavin earlier today, and his unprompted comment on the ∆Q framework for performance engineering is “our industry really needs this”. That endorsement speaks for itself.

Last chance to get early access to the ∆Q demo measurement system