Conference report Ultra Reliable Low Latency Communications 2017

I blagged my way into the rather good #URLLC2017 today. Here’s a quick synopsis of what’s going on, and my take on it, together with notes on each talk.

The telecoms industry is trying to progress beyond its two previous “smash hit” services: voice, and broadband Internet access. The two new key opportunities can roughly be characterised as “massive” (huge numbers of machine end points with sensors) and “critical” (low latency for interactive control). A third branch, “tactile”, for things like VR/AR, extends the current voice/broadband models.

The traditional way of working has been “horizontal”, with the same product being offered on the same basis to all users everywhere. These new URLLC opportunities are mostly centred on “verticals”, like automotive and healthcare, which means a total restructuring of the network and service offer.

The vertical ecosystem is different, the revenue model different, and the service SLAs to meet are different. 5G is seen as the primary technical means to deliver this new and valuable commercial end. The problem is that this raises a tectonic tension between different paradigms: “you get what you get” (historic supply-centric) and “you get what you need” (future demand-centric).

Telecoms executives have become used to delivering supply-centric products with either absolutely fixed timing and latency (at huge cost), or with variable timing and latency (at low cost). When it comes to these vertical uses, those models could be seen as “fit for everything” (if you can afford it) and “fit for nothing” (as the performance isn’t there and the risk too high).

The difficulty is now to get the best of both worlds: the benefits of resource sharing (packets, slices, virtual network functions, etc.) whilst also offering sufficiently predictable quality and performance, and simultaneously meeting tough-to-deliver new needs for security and automated operation. Easy, no?

At present, the 5G ambition seems to significantly exceed the ability of the industry to deliver. The people in the room are all individually competent and capable, but the systemic issues are presently “too big to succeed”. The whole industry lacks units for supply and demand, a common language for vertical requirements, and also needs to rebuild its fundamental protocols.

There is the potential for a “perfect storm”: the failure of performance integration, the tightening up of penalties for SLA breach, the arrival of new infrastructure competition (Facebook, Amazon, Alibaba), troubles of public safety systems on LTE, new high altitude and low-orbit satellite, and more.

There’s also a sense that the industrial sector, used to precision engineering, finds our performance engineering somewhat lacking in rigour. If the current telecoms industry won’t supply their needs, they may end up bypassing us and building a new industry on new science and architecture foundations. It’s already happened in some areas, like IoT.

This opens up the possibility for cloud players or industry specialists to disrupt the present mobile operator model. The defence of licensed spectrum may not be sufficient to prevent supply-side substitution, as regulators will listen to the needs of the demand side, as it employs more people and has more money.

The opportunity is very large – bigger than the original 1990s consumer mobility boom in many ways – so there’s plenty of scope for all to do well. Whilst telcos and their suppliers need to “up their game” to close the gap between demand and supply, it is achievable. The main change is attitude, not technology.

My sense is that the wild marketing claims and sexy use cases need to be toned down a bit. The real work is in defining reliability and low latency, creating interoperable SLA standards, calibrating scientific testbeds, and designing new business processes like automated fault isolation.

Not very glamorous, I know. But then again, 5G is the digital experience trucking industry, where the packets have no rest stops.

For additional event coverage, see the liveblog of Keith Dyer of TMN or try the conference hashtag #URLLC2017. For the full agenda and speaker bios see here.

Dr Mike Short, Telefonica – Conference Chairman

Demand side focus for industry
Fit-for-purpose as the theme
“Think about demand side as well as supply side, that would be great”

That Dr Short chooses to grace this event as Chairman says something about the subject’s perceived significance.

Adrian Scrase, CTO, ETSI

5G – town of 30k people – creates 300 jobs – $50m value
City like Chicago – 90k jobs
5G addresses new demand that 4G cannot reach
Mobile broadband not enough – also need massive M2M + URLLC
Need to talk in same terms across the industry
ITU defines the parameters
We “need to know what we mean for low latency” when we say “under 1ms”
1 million devices per sq km
<1 sec per day when cannot send 32 byte packet over radio path
Automotive, public safety, factory automation – each have their own requirements
NFV, MEC – mechanism responses from telecoms
“Today’s IP is not good enough. We need to do something better.”
Complex network slicing – hard to implement
Zero-touch network + service management – very new, many emerging activities
Police, Fire, Ambulance – broadband – groups of 5k people, device to device, new requirements
Mission critical video, data – not just blue light services (a small industry)
Chipsets have to be common – with multiple applications
LTE cannot do “massive” or “critical”.
Automotive – 10 years developing own standard
Huge scale, low component cost
High-speed trains – GSM-R – end of life
Looking to LTE+5G
Functional requirements similar to M2M; done gap analysis
Initially Europe, then global
Satellite – 3G poor, 4G missing. 5G new story – will be component, same radio but adjusted. Integral part of 5G system. Game changer – drones, high-altitude platforms.
Short timescales: new radio, new core network.
Don’t mention network management, too complicated for manual way. Automated and zero-touch. Will need AI to manage network.
Lacking – security, privacy, trust.
Careful system design needed to reach full potential.
5G is for everyone and everything – driven by industries and beneficiaries.
Agriculture, transport, automation – coming to us. Some missing.

5G is the answer to life, the universe, and everything. Or is that 42G?

Pierpaolo Marchese, Head of Standards Coordination, Telecom Italia

5G – complete new platform. Bigger journey – digital transformation. Cloudification, appification. New concepts like slicing. New skills.
Extended range of QoS values – core network
Service-based architecture – cloud-ready
State of the art protocols (HTTP2, RESTful)
Microservices architecture for core network
Slicing – multi-tenant, API driven
Attached without IP address allocation [RINA – association control]
LTE-M – low priority indicator for delay tolerant device
Distribute intelligence in network.
Use cases – robotics, cloud, UAVs, etc.
How to orchestrate and control slices in a distributed environment
Single protocol framework – SIM and non-SIM use cases
Federated network slicing
Next gen protocol stacks
Nanocore and AI – eSON [not going to work]
“The real voyage of discovery consists not in seeking new landscapes, but in having new eyes.” – Proust.

I wonder… what do all those acronyms mean? Can’t be that important.

Andy Sutton, Principal Network Architect, BT

Lots of low-latency requirements
Functional decomposition – RAN – one logical block [MG – what about non-functional decomposition?]
“CU pooling” “F1 flex” – [MG: I have no idea what they are.]
IPSec on F1 interface over Ethernet [MG: 2 out of 2 terms I grok.]
“How much latency can we afford?”
Optimise overall capacity and performance
AI + automation opportunity
“Performance” – [MG: Yay! Someone gets it.]
95^th percentile of distribution for latency
When to distribute content – CDN, MEC for VR/AR, compute for factory automation? [MG: How to do this? ∆Q!]
50% of content BT serves today is from on-net caches
Can have network-based low-latency services
Functional decomposition of RAN – whole host of new challenges.
Network very dynamic, NFV vital. Softwareisation of network critical for network.
URLLC is an overlay on base network of enhanced mobile broadband.
Constrained by imagination. [MG – no, science and maths.]
There is “UR” and “LLC” – not the same. Address independently.

I have no idea what any of the boxes do, but the arrows look like performance integration trouble.

Luis Lopes, Senior Staff Engineer, Qualcomm

Always-available, secure cloud access
Unified fabric – diverse services and devices
Trade-offs
Efficient multiplexing
“Working win partners looking for new use cases”

Engineering involves trade-offs.

Panel

Nokia – Handovers are hard in 5G
“What does it mean to deliver low-latency?”
Design goals – latency, reliability, energy, battery
Minimum Service Guarantee – SLA
Complex challenge – layers of definition – physical, synchronisation, placement of functions, etc.
Reliability cannot be disjoined from latency
<1ms – close to edge – under 150km
Mobility – makes latency and reliability requirement hard
Are we talking a few bytes or a few thousand bytes – requirement changes
Q: How are you going to define SLAs? A: “We don’t know”
Slice – minimum SLA attached to it
Throughput – latency – reliability – mobility
Time-sensitive networking – 802.1 committee
The network functions need to be placed close, else the radio effort is wasted
Lots of open questions and challenges
Should not fear cloud computing – need to cloudify network
Can then achieve performance, esp in terms of latency

Javan Erfanian, Bell Mobility

Lots of use cases (so can’t engineer them all individually)
Not just about capacity and coverage, but a wide range of requirements in a mobile + connected society
Dynamic reconfigurable environment
Service-based
Concurrent partitioned network-on-demand
Dynamic composition of network functions and distributed flows. [How?]
Can we get the 5G benefits without a 5G core?

Our industry is going from the bandwidth era to the latency era.

Panel

Key enabling tech? [MG framing issue – assumed problem is tech, but really is sci/maths]
5G – enabler for digital transformation
Need to move beyond mobile broadband
Expand ecosystem + value base
Need to move forward WITH verticals, not just assume. Something new for telecoms.
Need to engineer solutions for verticals that enable their business model change
When we started with small cells, every lamppost was seen as being made of gold; minset change from infrastructure to services being seen as valuable
Maturing as an industry, bigger societal impact
May be private networks, or virtual private networks, and less about public networks
Can’t have 4 operators on every lamp post
Network slicing – need a global agreement – same car sold everywhere, driven everywhere
There are gaps – defining slicing from vendor domain, but need to satisfy needs of verticals
Want network slices based on template, longer-term goal of fully automated delivery
4G has QoS definition, but not well used because of regulation
Need discussion about net neutrality, slicing has to deliver guarantees
Critical services for automotive – safety-based
These large industries not convinced 5G is a solution to their problems
Must demonstrate (live, not Powerpoint) value of 5G
Europe going from laggard to leader
Distributed self-optimising network is essential
5G is so different from 4G and 3G is needs a different funding model
The classic operator-vendor-human model won’t cut it; moving from serving humans to machines
New vertical-centric revenue model
Today’s operators are BT and Telenor, tomorrow’s could be Alibaba and Amazon

A preview of heresy to come. (“All IP” will soon get you fired.)

Dr Colin Wilcock, 5GIA

5GPPP – European research
5G Infrastructure Association
Thousands of researchers – hard to align
Trying to prove to verticals that 5G is the solution
Timescales – aggressive [unrealistic?]
Europe has lacked a focus event like the Olympics to drive cooperation and energy
Euro2020 soccer is one
Japan + Korea – one or two major operators, one or two vendors; Europe very fragmented in terms of countries, operators and vendors
Hard to agree common spectrum
Means we can try more things – and also do things like cross-border services
Forces us to learn interworking problems

I caught the end of ABB’s presentation. It’s all about the tail of the latency distribution, not its average.

Osman Yimaz, Team Leader, 5G Radio Resource Control, Ericsson

“Radio resource control” [MG: they haven’t grokked the distributed computing bit]
Ultra-lean – minimise transmissions (signalling + control overhead) [MG: Finally! Applying 1970s theory to telecoms, only 50 years late to party.]
Multi-service, multi-connectivity, multi-beam
Reduce size of radio slots
One-shot transmission vs many-short transmission vs retransmission
Single vs joint transmission (one vs many backhauls)
Need smarter scheduling

A better RAN is necessary, but not sufficient, to deliver 5G. The core problems are with the distributed computing architecture and its protocols.

Kevin Smith, Chair ETSI Next Generation Protocols, Vodafone

“We need to talk about TCP/IP”
Lots of great access technologies
Superhighways for legacy vehicles
Any flaws in protocol design increase processing
Workarounds – layers and shims – increase processing, latency, cost
More latency = less revenue
Parameters need to reflect current scenarios
Gone from 1970s – to 2010s – everything has changed
Network resource contention – huge number of flows
AMPS: Addressing – Mobility – Performance – Security
IP – interface-centic, not app-centric
Signalling storms
Performance – TCP congestion control not good, poor throughput for small flows, ridiculously large headers
Congestion bubbled up from all layers to the application – and a round trip
Need tighter control loops – at the right layer
Lots of wasteful retransmissions
TCP can’t tell layer 1 or 2 retransmit vs loss
Security – there is none. Need expensive overlays like IPSec. Well known ports are a vulnerability.
These are not solved by IPv6. Increases header size – wrong direction.
Semantic info of real estate – bind MAC address – vulnerable data. Anonymise it – more work. BGP becomes hard to scale in mobile
Score card for TCP/IP is bad. Going to get worse.
Future demands – huge change – quantity, quality, endpoints, network attachments
Just increasing capacity does not solve this problem
Industry verticals don’t just need 1ms radio, but end-to-end budget for performance
Four 9s reliability – a worry – and energy industry needs better
SDN, NFV slicing don’t solve this
Today’s protocols reduce the value of the network investment
For URLLC, shims and workarounds will break and fall apart
Alternatives live now
Narrowband IoT
Some secret ones
Next Gen Protocols – ETSI
Too much inertia to change Internet in one hit
Carrier Ethernet, slicing, multi-access edge computing are example applications
Temp shims to work with TCP/IP
Quality Attenuation methodology – can package up SLAs

If IPv6 is the answer, you are asking the wrong questions.

Dr Ilari Thibault, Principal Researcher, 5G Technologies, Vodafone

5GAA – 5G Automotive Association
Bring together automotive, telecoms, road operators
C-V2X – ability for all road users to cooperate
Translate service requirements from automotive into requirements for telecoms industry
High-density truck platooning
Cooperative adaptive cruise control
Varies a lot by choice of radio technology
Assumption – minimise inter-truck distance
Contextual constraints – e.g. road safety

Solving vertical industry problems means relating their business KPIs to your network ones.

Dr Andreas Muller, Corporate Research, Robert Bosch

5G expectation from manufacturing industry
Zero latency, errors, downtime, unlimited data rate, zero costs
Delivering Industry 4.0
Mobile and collaborative robots in factories – need to be wireless
Moving or rotating parts – wireless more attractive
Take out cables – remove breakage opportunity, lots of cable management
[Remote control to pres breaks – URLLC!]
Controller – [Set points] – Actuators – Process – Sensors – [Check actual values]
Using industrial Ethernet – TSN time-sensitive networking
1 microsecond synchronicity need
All about worst case latency, not average
Motion control hardest – safety traffic is “eight nines” requirement
Mobile energy efficiency – 10 years
Need just one network for the many uses cases; can’t run 10 networks
Local deployments, so don’t need nationwide
Lots of direct comms in a factory, no middle boxed/network
Cannot offload apps to an external cloud, but can bring cloud approach into factory to share resources
Security – should not be possible for external agent to jam the network
SIM-less identity system
Latency-optimised ciphers
Self-managing system
Who will operate a 5G network in a factory?
Exclusive access to spectrum needed – WiFi not viable
Huge concerns to let traditional MNOs inside factory
Don’t want the operational data to leak
Also locked in, economic issue
Liability issues, who is going to carry the risk? Millions of euros to recover from a production stoppage
Want to operate own network as a big company
Must have private networks
Want to have local spectrum allocations – geo region

Dr Andreas Muller of Robert Bosch setting some harsh requirements!

Steve Whatson, Deputy Director, Emergency Service Comms, The Home Office

TETRA is end of life
Unique coverage – take out site, lose coverage
Includes underground + air-to-ground
ESN – Emergency Services Network
3000 sites – old Airwave
19000 sites – new ESN
Priority and pre-emption of safety users against normal commercial users
Desire same resilience and coverage
Fewer generators on base stations, as can afford more to fail
Challenges – many departments funding, fragmented
Enormous implementation and transition project
Tradition period – working on both networks at the same time
Home Office is prime contractor – responsible for resolving issues
Performance isolation is core requirement
Safety case [backwards looking at past disasters]

Solving digital strategy problems for verticals means huge transformation projects

Prof Mischa Dohler, Kings College London

Interested in longhaul Internet with perceived low latency
Human VOR – 10ms
Can we communicate like that LA-London? No.
Can do some level of predictive control systems to cope with delay
Huge bandwidth challenge too – compression and encoding adds delay
Bandwidth predictions for 2025 wrong
Just 1% realtime causes 2 orders of magnitude error in capacity prediction
Need high bandwidth and low latency
Cooper’s Law
Most of gain in wireless capacity has come from smaller cells
Big bottleneck is not physical layer or spectrum
Architecture problem – fronthaul/backhaul
Unless nationalise/deregulate public street furniture, no way to deliver 5G
Kings College – first pre-release 15 5G call last week
All three radios running
Hybrid Access Gateway – routes traffic according to load; offload to WiFi/satellite/fixed
Hoping to place first national 5G call soon
Hospital to hospital comms
“Hospital on wheels”
URLLC will be a game changer
Cellular now useful to industry, not just consumers
Internet of skills – empower the people, don’t replace them

Something suspiciously similar to ∆Q’s G, S and V. It’s proper performance science!

Oussama Elhage, Specialist Registrar Urology, Guy’s and St Thomas’ NHS Trust

Zeus robotic system for remote surgery
Experimental still at this stage
Randomised controlled trial – establish utility of procedure
Needle into kidney
“The surgeon will Skype you now” – but limited by lag
Risk of hacking
Da Vinci surgical system current only robotic system in widespread use – knees and hips
Costs £1m-1.5m to buy, £100k/pa to maintain.
“An information system with arms is the 21^st century scalpel”
Like an expensive sewing machine
Additional 1-2 years of training to be signed off to do robotic surgery
A significant source of regret – high expectations and lack of touch feedback
Only visual clues, not tactile
Using iPad as AR visualisation of location of problem to overlay MRI scan on person
3D printed model of the organ from MRI scan to help surgeon know where to cut
Cost still a barrier; new system from Korea next year will help
Patient confidentiality also an issue with remote access
New security risks: who is responsible
Whose jurisdiction if patient in London is operated on remotely from Paris?

Dr Ilari Thibault, Principal Researcher, 5G Technologies, Vodafone [on NGMN activities]

Packet size, reliability, latency, coverage – fundamental trade-offs
If we use a TCP/IP stack, the headers use up all of our resources (lots of small packets, lighter stacks)
Otherwise the air interface is carrying nothing – just overhead

URLLC is not happening on TCP/IP because the bloated header is bigger than the payload a lot of the time. (I think you get the message: TCP/IP is not the future.)

Panel

Not clear to verticals the complexity of managing a network
If there’s an automotive or industrial accident, who is responsible?
Who will pay? Industry verticals already spending a lot of previous gen services.
Is 5G just carrier aggregation and MIMO antennas?
Government backing critical – money, demand, research

Chairman’s wrap-up