Carnegie
Mellon
FTTx Architectures and Why it Matters for the
Open Access Debate
Marvin A. Sirbu
Department of Engineering and Public Policy
Carnegie Mellon University
sirbu@cmu.edu
http://www.andrew.cmu.edu/user/sirbu/
© 2008 Marvin A. Sirbu
1
Carnegie
Mellon

FTTP networks have significant economies of scale




 facilities-based competition is unlikely to be
sustainable
Service-level competition can exist over shared
network infrastructure


Conclusions Up Front
Sharing possible at different levels
Sharing of dark fiber requires attention to fiber layout
There is great variety in the models of sharing which
can be found today
A wholesale-only provider is financially viable

It is not necessary to be vertically integrated to be
profitable
© 2008 Marvin A. Sirbu
2
Carnegie
Mellon




Outline
Models of Competition in FTTP
Alternative FTTP architectures: impact on competition
Economics of FTTP
Economics of a Wholesale/Retail split
© 2008 Marvin A. Sirbu
3
Carnegie
Mellon




Outline
Models of Competition in FTTP
Alternative FTTP architectures: impact on competition
Economics of FTTP
Economics of a Wholesale/Retail split
© 2008 Marvin A. Sirbu
4
Carnegie
Mellon
Facilities based competition – each
competitor builds FTTP network
Central Office s
Home 1
Service
Provider A
Network 1
Home 2
Separate Networks
Service
Provider B
Network 2
Data Link Layer Equipment
ATM, Gigabit Ethernet, SONET
© 2008 Marvin A. Sirbu
5
Carnegie
Mellon


UNE (LLU) based Competition in FTTP
Dark fiber based – network owner wholesales dark fiber
Wavelength based – network owner wholesales wavelengths
Central Office
Home 1
Service
Provider A
Network
Data Link Layer Equipment
Service
Provider B
Home 2
© 2008 Marvin A. Sirbu
6
Carnegie Open Access based competition – network owner
Mellon
wholesales transport capacity
Central Office
Home A
Service
Provider A
Network
Home B
Common Data Link
Layer Equipment
© 2008 Marvin A. Sirbu
Service
Provider B
7
Carnegie
Mellon
Sharing Network Infrastructure:
Summary
Layer:
Shared Infrastructure…
0
Conduit and collocation facilities.
1 (Physical Layer
Unbundling)
Dark fiber leasing, or perhaps, Optical Layer
unbundling (CWDM or DWDM in PONs)
2 (Data Link
Layer
Unbundling)
Dark fiber and link-layer electronics at each
end. For example, Ethernet-based VLAN,
or ATM-based PVCs.
3 (Network Layer
Unbundling)
Basic network service provided. For
example, IP Layer 3 service over cable
using policy-based routing to multiple
ISPs
© 2008 Marvin A. Sirbu
8
Carnegie
Mellon
0
Open access to ducts


1
Portugal
France
Dark fiber at layer 1

2
Examples of Sharing at Different Layers
Stokab in Stockholm
VLAN service at layer 2



UTOPIA
Amsterdam
Pau
© 2008 Marvin A. Sirbu
9
Carnegie
Mellon
Multiple Layer Separation Amsterdam
Source: http://www.citynet.nl/upload/Wholesale-bandwidth-Amsterdam-Citynet.pdf
© 2008 Marvin A. Sirbu
10
10
Carnegie
Mellon

If you build a wholesale network, will there be service
providers?


Kutztown, PA wanted to do only up to layer 2 and
couldn’t find service providers to run over the network
Operations finger pointing between wholesaler and
retailer


Issues and Problems
Provo Utah sold its layer 2 wholesale network to a service
retailer arguing that integrated operations are cheaper
Economies of scale

Operating company to light the fiber in multiple cities
– Axione
– Packet Front
© 2008 Marvin A. Sirbu
11
Carnegie
Mellon




Outline
Models of Competition in FTTP
Alternative FTTP architectures: impact on competition
Economics of FTTP
Economics of a Wholesale/Retail split
© 2008 Marvin A. Sirbu
12
Carnegie
Mellon
Home Run Architecture
ONU
Central Office Infrastructure

Implications for
Competition

Physical layer
unbundling
possible
– wholesaler can
sell individual fiber

Also supports open
access
Dedicated fiber to each Home
Central
Office
Equipment
OLT Port
ONU Optical Network Unit
OLT Optical Line Termination
Distribution
Loop
Feeder Loop
© 2008 Marvin A. Sirbu
13
Carnegie
Mellon
Active Star Architecture
ONU
Central Office Infrastructure
1

Implications for
Competition

Physical layer
unbundling is
difficult
Shared Feeder fiber
Central
Office
Equipment

OLT
N
Remote Node with
Active Electronics
Equipment


Distribution
Loop
Feeder Loop
© 2008 Marvin A. Sirbu
requires
competitors to
collocate
electronics at
remote node
Must provide
feeder fibers for
each competitor
Logical layer
unbundling possible
- supports open
access
14
Carnegie
Mellon
Curb side Passive Star Architecture (PON)
ONU
Central Office Infrastructure




Central
Office
Equipment
OLT
32

Distribution
Loop
Implications for
Competition

Physical layer
unbundling not
possible

Logical layer
unbundling
possible - supports
open access
Shared Feeder fiber
1


Curbside Passive
Splitter – Combiner
Separate λ’s may be used for
Data and video
Feeder Loop
© 2008 Marvin A. Sirbu
15
Carnegie
Mellon
WDM PON
ONU

Implications for
Competition

Physical layer
unbundling not
possible
Optical layer
unbundling
possible –
wholesaler can sell
wavelengths
Also supports open
access
Central Office Infrastructure

1
Shared Feeder fiber



Central
Office
Equipment

OLT
32

Passive Splitter –
Combiner


Distribution
Loop
Feeder Loop
© 2008 Marvin A. Sirbu
16
Carnegie
Mellon
Design Considerations in a PON: A
Curb-side PON
Neighborhood 1
Central Office Infra structure
Splitter 1
Central Office
OLT Equipment
PON1
Central Office
OLT Equipment
PON2
Splitter 2

Both OLTs needed if only one home in each splitter
group subscribes
Neighborhood 2
© 2008 Marvin A. Sirbu
17
Design Considerations in a PON:
A Fiber Aggregation Point (FAP) PON
Carnegie
Mellon
Neighborhood 1
Central Office Infrastructure
Splitter 1
Central Office
OLT Equipment
Splitter 2
Aggregation
Neighborhood 2

Fiber Aggregation Point PON supports all
models of competition
© 2008 Marvin A. Sirbu
18
Carnegie
Mellon
How many homes should be aggregated
at an Optimal FAP?
FTTH Costs for an Urban Deployment
NPV of Cost per Home
1600
1500
Home Run
1400
Active Star
1300
PON
1200
Distributed
Split PON
Home Run
PON
1100
1000
900
800
0
200
400
600
800
1,000
Num ber of Hom es Aggregated at OFAP

OFAP allows deferring investment in OLTs until
penetration requires it
© 2008 Marvin A. Sirbu
19
OFAP as a Real Option to
Phase-in New Technologies
Carnegie
Mellon
Neighborhood 1
Central Office Infrastructure
Splitter 1
GPON
GPON CO OLT
Equipment
BPON CO OLT
Equipment
Splitter 1
BPON
Neighborhood 2
Aggregation
© 2008 Marvin A. Sirbu
•OFAP also supports flexibility
in future split ratios
- 10 Gbps GPON, GEPON
- WDM PONs
20
Carnegie
Mellon
OFAP Benefits with
an Active Star Architecture
Neighborhood 1
Central Office Infrastructure
Feeder 1
Central Office
OLT Equipment
RT & OLT to
be deployed as
needed
Central Office
OLT Equipment
Feeder 2
Neighborhood 2
•Larger serving area
Aggregation point
•Higher utilization of RT and OLT ports
•Neighboring homes can be served by different
technology generations
© 2008 Marvin A. Sirbu
21
Carnegie
Mellon



As video becomes dominated by unicast Video on Demand
(VOD) metro aggregation network costs soar
In smaller communities, access to regional transport to a Tier
1 ISP is a major barrier to entry
Retail service providers sharing an FTTH access network
may also need to share at the metro/regional level in order to
be economically viable.


Sharing in the “Second Mile”
NOAAnet
There is a tradeoff with distributed video servers

Sharing a content delivery network (e.g. Akamai) may be an
alternative.
– This requires distributed colo space and interconnection

See Han, S. et al “IPTV Transport Architecture Alternatives and Economic
Considerations,” IEEE Comm Mag, Feb 2008
Lamb L., “The Future of FTTH – Matching Technology to the Market in the
Central Office and Metro Network,” NOC 2008.
NSP, “A Business Case Comparison of Carrier Ethernet Designs for Triple
Play Networks,”


© 2008 Marvin A. Sirbu
22
Carnegie
Mellon

If regulators want to be able to require dark fiber
unbundling, they need to require compatible fiber
layout



Regulatory Implications
OFAP PON vs curb-side PON
Even larger OFAP for competitive active star
– Need for additional feeder fibers for competitors
All architectures support logical layer (“bitstream”)
unbundling


IPTV unbundling possible at bitstream layer
If video distributed over a separate wavelength, issues of
access to RF multiplex.
© 2008 Marvin A. Sirbu
23
Carnegie
Mellon




Outline
Models of Competition in FTTP
Alternative FTTP architectures: impact on competition
Economics of FTTP
Economics of a Wholesale/Retail split
© 2008 Marvin A. Sirbu
24
Simple FTTH Economics:
FTTH Includes Fixed Plus Variable Costs
Carnegie
Mellon
Cost = Fixed +
R * Variable
$

e.g. for Verizon YE06


Fixed=$850
Variable=$880
Source:
http://investor.verizon.com/news/20060927/200
60927.pdf
Fixed
costs
0%
100%
Take Rate
(R = customers / homes passed)
Adapted from Friogo, et.al.
http://ieeexplore.ieee.org/iel5/35/29269/01321382.pdf
© 2008 Marvin A. Sirbu
25
Carnegie
Mellon
Cost Per Subscriber vs Take Rate
Variable Cost
Total Cost/Sub
8000
Cost/Subscriber
7000
6000
5000
4000
3000
$1730
2000
1000
0
0%
20%
40%
60%
80%
100%
Take Rate
© 2008 Marvin A. Sirbu
26
Carnegie
Mellon

One operator estimates $90/month per subscriber






$40 for ongoing services cost
 $50/month to cover capital costs
Assume an average of 10 year lifetime, 5% cost of
capital


How Much Revenue to Support FTTH?
Fiber lasts 40 years
Electronics lasts five years
$50/month can amortize $4700
What if Average Revenue Per User (ARPU) is less?
$30/month can amortize $2800
© 2008 Marvin A. Sirbu
27
Carnegie
Mellon
Cost Per Subscriber vs Take Rate
Variable Cost
Total Cost/Sub
Capital at $50/mo
Capital at $30/mo
8000
Percent take rate needed to break even
7000
Cost/Subscriber
6000
Capital that can be amortized with $50/mo/sub
5000
4000
Capital at $30/mo/sub
3000
2000
1000
0
0
Adapted from Frigo et. al.
0.2
0.4
0.6
Take Rate
© 2008 Marvin A. Sirbu
0.8
1
28
Carnegie
Mellon
Cost Per Subscriber vs Take Rate
8000
Take Rate
7000
Consumers
Cost/Subscriber
6000
Capital that can be amortized with $50/mo/sub
5000
4000
3000
2000
Competition
1000
0
0
Adapted from Frigo et. al.
0.2
0.4
© 2008 Marvin A.Take
Sirbu
0.6
Rate
0.8
1
29
Carnegie
Mellon



Economic Implications:
If revenue available to amortize plant is only
$30/month, must reach penetration of > 45%
 room for at most 2 facilities-based providers
This analysis understates the problem


No customer acquisition (marketing/sales) cost included
– Customer acquisition drives up Fixed costs pushing
breakeven penetration higher
Unlikely to see >90% total penetration
© 2008 Marvin A. Sirbu
30
Carnegie
Mellon

Facilities-based competition among fiber network
providers is unlikely


Economies of scale
Regulators should be cautious of waiving open access
requirements in return for investment in fiber


Regulatory Implications
Could lead to remonopolization
At best duopoly competition



If service competition limited to ISPs which own facilities
greatly reduced service level competition
Operators will have Significant Market Power (SMP)
Reduced service-level competition raises Network
Neutrality issue
© 2008 Marvin A. Sirbu
31
Carnegie
Mellon

Net Neutrality
Can third parties compete with vertically Integrated
ISPs?
Central Offices
Home 1
Service
Provider A
Network 1
Home 2
Apps
+
Content
Separate Networks
Service
Provider B
Network 2
Apps
+
Content
Data Link Layer Equipment
© 2008 Marvin A. Sirbu
32
Apps
+
Content
Carnegie
Mellon




Outline
Models of Competition in FTTP
Alternative FTTP architectures: impact on competition
Economics of FTTP
Economics of a Wholesale/Retail split
© 2008 Marvin A. Sirbu
33
Carnegie
Mellon

Open Access: Network operator provides wholesale
transport to service providers


Economic Analysis: Motivating Question
Do sustainable prices exist for an infrastructure-only
provider?
Build a supply/demand model and calculate welfare
effects for different industry structure models
Central Office
Home A
Service
Provider A
Network
Home B
Common Data Link
Layer Equipment
© 2008 Marvin A. Sirbu
Service
Provider B
34
Carnegie
Mellon

Structural separation interferes with the
ability to price discriminate
Vertically integrated entity
‡ Can sell 7 bundles: Voice, Data,
Video, Voice-Data, Voice-Video, DataVideo, Voice-Video-Data
‡ Can set 7 prices


Does this make
a wholesaler
less likely to
recover costs
vis-à-vis a
vertically
integrated
entity?
Dark fiber wholesaler
‡ Can sell only dark fiber access
‡ Can set only one price
© 2008 Marvin A. Sirbu
35
Carnegie
Mellon


Wholesale Prices and Arbitrage
A dark fiber wholesaler can set only one price
A lit fiber wholesaler can set a price for data or video
bandwidth but cannot set a separate price for the
bundle


Video bandwidth is sufficient to offer both video and data
services to customers, so
Wholesale price of “bundle” bandwidth and “video”
bandwidth must be the same
© 2008 Marvin A. Sirbu
36
Carnegie
Mellon
We have studied 3 models
Assumptions
FTTP

Single Service
Provider
2-service
network only network serving market
Voice services are provided over a separate network
FTTP network used to provide only data and video services
FTTP

Single Service
provider
3-service
network only network serving market
FTTP network used to provide voice, video and data service
Market

Duopoly
2-service
already served by (cable) incumbent when FTTP
provider enters
FTTP and incumbent network used to provide only data and
video services
© 2008 Marvin A. Sirbu
37
Carnegie
Mellon

Two-service model for the
Wholesale-Retail Split
Demand Model

Consumers have different willingness to pay for voice, video and
data services: Willingness to pay for a particular service can be
modeled by a statistical distribution for a particular market

There is correlation between the willingness to pay for voice,
video and data for one particular consumer: One can imagine a
3-space where the coordinates of each point give her willingness
to pay for voice, video and data services

For simplicity, here we assume everyone wants voice – so our
demand model is 2-space, where the coordinates of each point
give the willingness to pay for data and video
© 2008 Marvin A. Sirbu
38
Demand Model..
Carnegie
Mellon
X1=Homes taking service1 (data) at price P1 (Area BDP1P3)
X2=Homes taking service2 (video) at price P2 (Area ACP2P3)
X3=Homes taking service3 (video and data) at price P3 (Area ACDBZ)
Willingness to Pay
A
160
Z
140
120
Video
P3
100
80
P2
60
C
40
B
D
20
0
0
20
40
P1
60
80
Data
© 2008 Marvin A. Sirbu
P
1003
120
140
39
Carnegie
Mellon




Supply Model
Annualized Fixed cost for wiring up the entire market consisting of X
homes = F
Annualized Fixed Cost of installing CPE and drop loop = C0
Annual incremental cost of providing data service (Service 1) per
home = C1
Annual incremental cost of providing video service (Service 2) per
home = C2

Observation: Marginal Cost of Bundle (C0 +C1+C2) is less than the sum
of Marginal Cost of Data (C0 +C1) and Marginal Cost of Video(C0 +C2)

If X1 homes take data service, X2 homes take video service and X3 take
both, annual cost of providing service =
F + C0(X1+X2+X3) + C1X1 + C2X2 + (C1 +C2)X3
© 2008 Marvin A. Sirbu
40
Carnegie
Mellon

Vertically Integrated entity (Network owner provides retail
service)



Possible Industry Structures
‘Verizon’ Model (Profit Maximizing)
‘Bristol’ Model (Welfare Maximizing)
Structurally Separated entities (Network owner, either by
regulation or choice, is only a wholesaler. The retail market is
assumed to be competitive/contestable)




‘Grant County Profit (GCP)’ (Profit Maximizing layer 2 service
wholesaler)
‘Grant County Welfare (GCW)’ (Welfare Maximizing layer 2
service wholesaler)
‘Stockholm Profit (SP)’ Model (Profit Maximizing dark fiber
wholesaler)
‘Stockholm Welfare (SW)’ Model (Welfare Maximizing dark fiber
wholesaler)
© 2008 Marvin A. Sirbu
41
Carnegie
Mellon


Model Results
Not surprisingly, if network owner optimizes Social
Welfare (e.g. Bristol) consumers are much better off
than if network owner optimizes profit
If network owner optimizes profit, THERE IS
VIRTUALLY NO DIFFERENCE in profit for a vertically
integrated firm or a wholesaler.


The fact that vertically integrated firm has more flexibility
to price discriminate is not important since most
households subscribe to the bundle, and wholesaler can
extract the same rent.
If there is a large fraction of the population with no
interest in broadband data, then vertically integrated firm
can do 25% better than a dark fiber wholesaler, but still
no better than a lit fiber wholesaler.
© 2008 Marvin A. Sirbu
42
Carnegie
Mellon
3 services model shows less than 5%
difference Stockholm and Verizon profits
F=5x104
C0=8
C1=20
C2=30
C3=5
1= 35
σ1= 10
2 = 45
σ2 = 10
3= 25
σ3= 10
© 2008 Marvin A. Sirbu
43
Carnegie
Mellon
Similar profits are attained in spite of a
different distribution of subscribers
© 2008 Marvin A. Sirbu
44
Carnegie
Mellon

If services are identical, classic case of natural
monopoly


Firm with higher penetration has lower costs
Ruinous competition



What if There Are Competing
FTTP Operators?
Having sunk cost in fixed plant, each competitor is willing
to price at marginal cost
 negative profits
Stable competition can exist only if there are


Differentiated services appealing to heterogeneous
customer tastes; or
High switching costs
© 2008 Marvin A. Sirbu
45
Carnegie
Mellon




Duopoly Model Results
We assume two operators with similar cost structures,
one an incumbent, one a new entrant
Assuming video and data services are sufficiently
differentiated between competitors, both can survive in
the marketplace
If the new entrant is a wholesaler only, or vertically
integrated makes no difference in its profit
An incumbent competing against a dark fiber
wholesaler is modestly worse off than when competing
against a vertically integrated competitor

Wholesaler’s inability to price discriminate forces
competitor to reduce price discrimination and lose profit.
© 2008 Marvin A. Sirbu
46
Carnegie
Mellon
Model assumptions and caveats

Retail industry assumed to be perfectly competitive and no
entry barriers; retailers make zero economic profit

Revenues derived entirely from end customers, not from
application service providers

No economies of scope at retail assumed

Incremental costs, Ci , are the same in both vertically
integrated and competitive retail cases


Competition should drive down incremental costs of services
Layer 2 costs, C0, are the same whether supplied
competitively or by wholesaler

See above
© 2008 Marvin A. Sirbu
47
Carnegie
Mellon

Regulatory Policy Implications
Operators, municipalities or communities that build out
FTTP and choose to be wholesalers:
‡ (i) can realize sustainable prices,
‡ (ii) are likely to create greater welfare (due to innovation
spurred by retail competition) and
‡ (iii) are just as likely to recover costs (vis-à-vis vertically
integrated entities)

Model results contradict claims by operators that
vertical integration is necessary to support investment
in FTTP infrastructure
 regulatory holiday for FTTP investment is unwarranted.
© 2008 Marvin A. Sirbu
48
Carnegie
Mellon

What are the different models of competition in FTTP?




OFAP supports fiber unbundling even for PONs
More feeder fibers required for competition
FTTP networks have significant economies of scale



Facilities based
Service level (over shared network infrastructure)
Fiber layout affects options for competition


Conclusion
Unlikely to support multiple facilities-based providers
“Second Mile” sharing also important
A Wholesale Operator can earn profits similar to those
available to vertically integrated competitors

It is not necessary to be vertically integrated in order to
“earn enough” to pay for the infrastructure
© 2008 Marvin A. Sirbu
49
Carnegie
Mellon
For Further Information

http://www.andrew.cmu.edu/user/sirbu/pubs/Banerjee_Sirbu.
pdf

http://web.si.umich.edu/tprc/papers/2006/648/Banerjee_Sirbu%20TP
RC_2006.pdf
http://cfp.mit.edu/groups/broadband/muni_bb_pp.html

© 2008 Marvin A. Sirbu
50