Pages

Pages

Wednesday, September 07, 2016

5G: Will it be Sliced... or Hacked?

I've been giving a lot of thought recently to 5G - the technology, major use-cases, likely business models, timelines and implications for adjacent sectors such as IoT.

5G fits into both my own TelcoFuturism analyst/advisory work on the intersections of multiple technologies in telecoms, and also my secondary role working with STL Partners as Associate Director and lead analyst of its Future of the Network research programme (link). 

A philosophical split is emerging among operators and vendors:
  •  "One network to rule them all" idealists
  • "Make it functional ASAP & add other stuff later" pragmatists
There are various nuances, middle-ground thinkers and shades of grey, but in general the former tend to be companies driven by the core and services domains, and the latter have a radio/access bias.

The core-network group tends to view things through the lens of NFV, and with a 2020 target date. It sees a world that spans diverse 5G use-cases from smartphones to sensors to vehicle-to-vehicle communications, taking in police cars and replacing FTTH and WiFi along the way. It wants to use sophisticated MANO (management and orchestration) layers and next-gen OSS to create network "slices", supposedly from "end to end". Such slices would, in theory, be optimised for different business models, or verticals, or virtual networks - leaning heavily on policy-management, differentiated QoS and assorted other big service-layer machinery. Mobile edge-computing would, ideally, extend the operator's cloud infrastructure into a distributed, Amazon-beating "fog". Often, terms like "HetNet" will be added in, with the notion that 5G can absorb (and assimilate) WiFi, LPWAM, corporate networks and anything else into a unified service-based fabric.

The other group is driven by more pragmatic concerns - "better faster cheaper" upgrades to 4G in a 2018-19 timeframe (and 2017 trials), replacing DSL in rural areas where fibre is too expensive but cable is growing, more spectral efficiency to squeeze more usage out of frequency allocations, lower-cost mobile broadband for emerging markets, better cell-edge coverage, and (ideally) lower power consumption for the RAN. Perhaps unsurpringly, they focus more on the nuts and bolts of radio propagation in different bands, different modulation mechanisms, frame structures needed to optimise latencies - as well as practicalities such as small-cell backhaul. Business model discussions are secondary - or at least decoupled - although obviously there is a large IoT element again. The core network may well remain the same for some time, and 5G access will not necessarily imply NFV/SDN deployment as a pre-requisite. (I've spoken with CTOs who are quite happy without virtualisation any time soon).

In my view, it is the latter group which better understand the "hard" technology compromises that need to be made, as well as the timing considerations around deployment, spectrum availability - and the implied competition from diverse substitute technologies like SigFox, gigabit-speed cable, near-ubiquitous WiFi and even next-gen satellite (assuming no more SpaceX Falcons have unfortunate "anomalies"). A key concern is how to squeeze the ultra-low latency capabilities into a network architecture that also supports low-cost, mass IoT deployment.

Conversely, the other camp is often guilty of wishful-thinking. "Let's control flying public-safety robots with millisecond latency & QoS via MEC nodes & 6GHz+ licensed-band 5G from totally virtualised & sliced service creation & activation platforms". This would probably work as the basis of a 2023 Michael Bay movie, but faces quite a few obstacles as a near-term mobile operator strategy. [Note: this concept is only a very mild exaggeration of some of the things I've had suggested to me by 5G zealots].




There are various practical and technical issues that limit the sci-fi visions coming true, but I want to just note a couple of them here:
  • It is far from clear that there will be enough ultra-performance end-points to justify having the millisecond-latency tail wag the 5G dog. I'd guesstimate a realistic 100 mllion-or so device target, out of a universe of 10-20bn connections. Unless the related ARPU is huge (and margin after all the costly QoS/slicing gubbins is added in), it's not justifiable if it delays the wider market or adds extra complexity. Given that 100m would also likely be thin-sliced further with vertical-specific requirements (cars, emergency, medical, drones, machinery etc.) the scale argument looks even weaker.
  • A significant brake on NFV at the moment is the availability of well-trained professionals and developers. As one telco exec put it to me recently "we don't have the resources to make the architects' dreams come true". And this is for current NFV uses and architectures. Now consider the multi-way interdependencies between NFV + 5G + verticals + Cloud/MEC. The chances of telcos and their vendors building large and capable "5G slice" teams rapidly are very small. What would a "5G Slice development kit" look like? How exactly would an IoT specialist create a 5G-enabled robot anti-collision system for a manufacturing plant with arc-welders generating radio interference, for example? (And let's leave aside for now the question of what 5G NFV slices look like to regulators concerned about neutrality....)
In other words, I think that the "slice" concept is being over-hyped. It sounds great in principle, but it's being driven by the same core-network folk who've been trying to sell "differentiated QoS" in mobile for 15+ years. It took 7+ years to even get zero value-add VoLTE phone calls to work well on 4G with QoS, when that service had been specced and defined to within an inch of its life by committee. The convenient IoT/NFV/5G developer SDK and corresponding QoSPaaS isn't appearing any time soon.

That said, I'm sure that there will be some basic forms of network-slicing that appear - perhaps for the public-safety networks that are moving to 4G/5G whether it's truly ready or not. But the vision of 10, 100 or 1000 differentiated 5G slices all working as a nicely-oiled and orchestrated machine is for the birds.

Instead, I think the right metaphor is hacking not slicing. I don't mean hack in the malware/blackhat-in-a-basement sense, but in terms of taking one bit of technology and tuning/customising it and creating derivatives to serve specific purposes.

We already see this with 4G. There's a mainstream version of LTE (and LTE-A and enhancements), but there's also PS-LTE for public safety, NB-IoT for low-end IoT, and LTE-U/LAA/MuLTEfire for unlicensed spectrum. Those are essentially "hacks" - they're quite different in important ways. They benefit from the LTE mother-spec's scale effects and maturity - and probably would not have evolved as standalone concepts without it. In a way, the original railway GSM-R version of GSM was a similar hack.


I think 5G will need something similar. As well as the so-called "New Radio", there is also work being down on a nextgen core - but there may well also have to be spin-off variants and hacks as well. This could allows the mainstream technology to avoid some possibly-intractable compromises, and could also be a way to bring in vertical specialists that currently think the mobile industry doesn't "get" their requirements - as per my recent post that telecoms can't just be left to the telcos (link). 

As usual, the biggest risk to the mobile industry is strategic over-reach. If it persists in trying to define 5G as an all-encompassing monolithic architecture, with the hope of replacing all fixed and private networks, it will fail. The risk is that if it tries to create a jack-of-all-trades, it will likely end up as master-of-none. 5G has huge potential - but also needs a dose of pragmatism, given that it is running alongside a variety of adjacent technologies that look like potential disruptors.

Ignore the sneers that SigFox is just 2016-era WiMAX
and look at the ever-present use of 3rd-party WiFi as a signpost - and the emergence of WiGig. Look too at the threat that SD-WAN is having against MPLS and NFV-powered NaaS in fixed-line enterprise networks (link) - an illustration of the power of software in subverting telco-standardised business models. This time around, non-3GPP wireless is "serious" - especially where it leans on IEEE and ethernet.

In its fullest version, the "slice" concept is far too grandiose and classically 1990s-era telco-esque. Hacking is much more Internet/IETF-style "rough consensus and running code". It will win.


A forthcoming report on the Roadmap for 5G will be published as part of STL Partners' research stream on the Future of the Network soon. Please contact STL or myself (information AT disruptive-analysis DOT com) for more details, or for inquiries about custom advisory work and speaking engagements on 5G, NFV, LPWAN and related themes.

No comments:

Post a Comment