2. Consultancy on the future of telecoms.
Business model innovation.
Technology & product ideation.
Organisation development.
Public & private workshops.
I work with smart people on the hard stuff.
14. 2010s & 2020s
Packet-based
statistical multiplexing
(using IP)
Time-division
multiplexing
Best of
both worlds!
1980s and 1990s
2000s & 2010s
These changes are intimately
tied to our technology
choices, and how we share
the network’s fixed and
finite resource.
16. Meet Alice and Bob
It’s the end users who put the money into the system, and
decide whether we are creating value worth paying for.
Alice and Bob are end users.
17. Alice is a sales woman
Alice Smith
Sales Director
alice@widgetcorp.com
555-125-9876
18. Bob is her customer
Mustn’t
run out of
widgets!
32. TELCOSEND USER
Insurance
Second carINSURANCE Contingency fund (lawsuit, PR)
If Alice can’t rely on her communications service
delivering a good enough experience, she needs a
Plan B – and has to carry all the associated costs. For
a corporate user, that could be having multiple access
providers; for a consumer, you need to buy a DVD in
case Netflix doesn’t work.
Telcos spend a lot of
money dealing with
screw-ups.
33. TELCOEND USER
The whole picture
BENEFIT
COST
RISK
(failed call)
Made the sales call
Price of phone call
Didn’t make sale
Second car
Revenue
Tin, opex
SLA breach or churn
Unplanned capacity upgrade, fines
Time wasted
Reputational loss
INSURANCE Contingency fund (lawsuit, PR)
Frustration
Lots of (potentially
hidden) costs!
34. Who profits and why?
CUSTOMER SUPPLIERWHOLESALER
BENEFIT
COST
RISK
END
USER
END
USER
INSURANCE
Who &
how big?
35. How is the role of
wholesale changing?
BENEFIT
COST
RISK
INSURANCE
Voice Broadband Cloud services
39. Retail telco risk passed to
wholesaler
Retail telco only pays wholesaler
for a successful phone call
$
40. Wholesale telco risk is transferred
Wholesaler only pays terminating network
for a successful phone call
$
41. The delivery technology was
(and often still is) TDM circuits.
No overlaps or collisions!
This isolated flows and assured end user
outcomes – audible voice calls.
42. Composable value chain with SLAs
Retail
origination
carrier
Retail
termination
carrier
Wholesale
delivery
carrier
Predictable application outcome
43. How did we do?
Retail
origination
carrier
Retail
termination
carrier
Wholesale
delivery
carrier
49. VOICE SUMMARY
Resource efficiency
Flowefficiency
LOW HIGH
LOW
HIGH TDM
(core)
TDM
(edge)
Very profitable
wholesale business!
How hot can
we run the
network?
How well can we
deliver the flows
within acceptable
loss and delay?
60. End user risk
Alice bought ‘bandwidth’
(speed and mechanisms like 4G and FTTH)
$
61. ISP is blind to Alice’s needs
No visibility of application outcomes
$
62. Retail telco risk passed to
wholesaler
Retail telco buys ‘bandwidth’ transit.
Has no means of expressing
needs for application outcomes.
$
63. Wholesale telco risk is transferred
Wholesaler connects to receiving ISP,
but is carrying effective risk
of application failure
$
64. And what about Bob?
His ISP has no visibility of whether the
service is delivering a good outcome.
65. What are the risks?
Retail
origination
carrier
Retail
termination
carrier
Wholesale
delivery
carrier
QoE
RISK
QoE
RISK
PLAN B PLAN B
66. What are the risks?
Retail
origination
carrier
Retail
termination
carrier
Wholesale
delivery
carrier
Churn Churn
67. What are the risks?
Retail
origination
carrier
Retail
termination
carrier
Wholesale
delivery
carrier
Capacity
upgrade
risk
Capacity
upgrade
risk
68. And the wholesaler?
Incentivised to find cheapest means
of delivering ‘bandwidth’,
regardless of impact on user or rest of supply chain
69. Current wholesale data business is
structured in an unhelpful way.
1.Has become
disconnected from the
value end users are
seeking.
2.This misallocates cost
and risk.
71. There price we had to pay…
Time-division
multiplexed
circuits
Packet-based
statistical
multiplexing
TELEPHONY BROADBAND
Game of
chance
was easy
Game of
chance is
hard
91. Operator cost risks
Our experience is that these
hazards are poorly modelled in
broadband networks (if at all)
SLA breach
or churn
Unplanned
capacity upgrade
92. Summary so far
BENEFIT COST RISK
1. There is value in
effective task substitution
2. There are associated
benefits, costs & risks
3. It matters how big these
are, and how they are
allocated
97. Summary so far
Failed call
Plan B
SLA breach or churn
Unplanned capacity upgrade
TELCOSEND USER
Resource efficiency
Flowefficiency
LOW HIGH
LOW
HIGH
HELL
HEAVEN
1. Networking is a statistical
game of chance
2. That schedules resources
and determines
application performance
3. That controls the QoE and
cost hazards for both the
user and network operator
98. How can we get the best
of TDM and IP?
1. How can we increase benefits
(substitute for more valuable tasks)?
2. How can we lower costs by running
networks hotter?
3. How can we quantify, mitigate and
allocate hazards appropriately?
100. The cloud economy
Different demands to
those of voice and
today's broadband.
Understanding the nature of future demand, and how to create
appropriate supply, is the key to future success. People want to achieve
higher-value task substitutions: teleworking, education, healthcare. These
require dependable networks and application outcomes.
101. What has to change?
NOW FUTURE
PURPOSE-FOR-
FITNESS
FITNESS-FOR-
PURPOSE
107. What has to change?
NOW FUTURE
MONOSERVICE
NETWORKS
Weak
exploitation of
stat mux trades
POLYSERVICE
NETWORKS
Strong
exploitation of
stat mux trades
108. Multiple classes of service
Economy Standard Premium
Airlines offer multiple tiers of service, based on comfort, rather than arrival
time (bar a bit of queue-jumping at check-in). In networks, the tiers of
service are based on how much you can time-shift traffic.
118. Example services:
Bulk content cost reduction
$
$
$
BULK DATA DELIVERY
Bulk content providers are paid to mark time-shiftable traffic
to reduce peak loads and network cost for ISPs. Wholesalers
are ‘market makers’.
119. Example services:
Small cell assured backhaul
ASSURED BACKHAUL
$ $
3GPP standards were all built for a TDM world. Things are not working out that well
moving to IP. Quality-assured IP backhaul would restore the necessary properties for
small cells to work well.
125. Think logistics, not transport
It’s all about the trading space, matching supply to demand, offering a variety of
quantities, qualities and cost, and doing it over multiple timescales.
Small cellsHome workingEducationOTT voice and video delivery partnerships
I have four simple messages from this presentation
We will look at some of the potential service offerings.
Every network access involves creating value by substituting for something less beneficial or more costlyThe alternative could be another online or digital experience; it does not have to be offlineDoing it well means making good experiences common, and bad ones sufficiently rare.Asymmetry between success modes and failure modes.Alice defines the whole value of this substitution; not the operator
Excess risk has to be (self-)insured
TDM for voice had highly-utilised network cores, and low peak-to-mean at the network edge
TDM for voice had highly-utilised network cores, and low peak-to-mean at the network edge
TDM for voice had highly-utilised network cores, and low peak-to-mean at the network edgeEveryone made money!
However, TDM is not a suitable basis for delivering bursty data applications, because it leaves the network empty most of the time. When we switched to packet data broadband services, we gained resource efficiency through statistical multiplexing gain.
If it doesn’t work, Bob complains or churns.
TDM for voice had highly-utilised network cores, and low peak-to-mean at the network edge
at the network edge, where there is high contention between flows.
We will look at some of the potential service offerings.
TDM for voice had highly-utilised network cores, and low peak-to-mean at the network edge
Run the network hot, and deliver good experiences
TDM for voice had highly-utilised network cores, and low peak-to-mean at the network edge
TDM for voice had highly-utilised network cores, and low peak-to-mean at the network edge