Pricing Communication Networks -- an overview

1
Pricing Communication Networks-- an overview

• Content from Chapter 1-4 (Part A) of the book
• Costas Courcoubetis, Richard Weber, Pricing
  Communication Networks Economics, Technology
  and Modeling. Wiley, 2003

2
Preface

• (Traditional) engineers develop communication
  services without reference to how they should be
  priced
• Advance of technology has created a new and
  competitive environment for communication service
  providers
• Pricing and Competition issues are worthy of
  study, because
• - pricing affects how service is provided, and
  how resource is consumed
• - pricing mechanism provides incentives to
  control performance and increase stability
• - pricing mechanism allows more flexible service
  and efficient resource usage
• - Competition and regulation issues are
  important especially for control of bottleneck
  resource

3
Preface
Communication service
price
shape
Traffic
Meat of this book - generic economic models
that allow the network service to be priced
(like any other goods) - solution of fairness and
resource allocation problems based on pricing -
theoretical framework to price contracts (with
dynamic and negotiable parameters) - Review of
current research topics (incentives,
regulation...)
4
Preface

• This talk with cover
• the general picture (the market, role of price
  and
• economics, etc.) (chapter 1,2)
• preliminary modeling pricing a single link
  (chapter 1 4)

• Useful tips...
• Quality affects price
• Higher price, Lower demand
• Rule of market more competition leads to lower
  price
• Differentiate network service from traditional
  goods

5
Outline

• Communication service
• - Characteristics
• - Development of the market
• - Role of economics
• Generic Questions
• Tariffs and Contracts
• Pricing a single link

6
Communication service- Characteristics

• how to evaluate the value of the service?
• network externality the more to connect, the
  higher value
• Metcalfes law n(n-1)/2
• Andrew Odlyzko nlog(n)

More sold, more value for network
More sold, less value for traditional goods
There lies a danger for overdesigning the network
7
Communication service- Characteristics

• High construction (fixed) cost, but very low
  operating (marginal) cost

Rule of market competition drives the price
towards marginal cost (danger the price for
communication may approach zero!)
Definition Marginal cost (MC) is the change in
total cost (TC) when the quantity (Q) produced
increases by one unit, i.e.
MCdTC/dQ
8
Communication service- Characteristics

• Compare with traditional economic goods
• - Similarity decrease price -gt more demand
• increase price -gt less
  demand
• - Analogy view a network as a factory that can
  produce various combinations of network services,
  subject to technological constraints on the
  quantities of simultaneously provided services.

A factory whose capacity is limited by the
quantity of simultaneously provided products
9
Communication service- Characteristics

• Cheap or pricy? Avoid commodity!

1) Websites (like Yahoo, Sina...) can not charge
users for reading the news, because there are
hundreds of other websites offering exactly the
same information. The market is commoditized.
2) MS Word sells at good price because - it
is hard to learn another word processing
software - there are many other people using
the same software (network externality!)
3) Currently, the long-haul bandwidth (core
network) market has been commoditized (due
to dot-com bubble).
10
Communication service- Characteristics

• Special features

- No unified expression for quality (peak rate,
delay, loss rate, jitter, ...)
- User can buy one service (which is intended for
purpose A) for purpose B as long as the quality
level is adequate (substitutability)
- User can buy one service, and create two
services from it (splitting)
11
Communication service- Characteristics

• Special features

- Statistical multiplexing, i.e., certain level
of overbooking is allowed. e.g. two users on a
link, one with a peak traffic rate of
1Mbps, and a mean rate of 0.5Mbps, another
with a peak traffic rate of 1Mbps, and a mean
rate of 0.6Mbps, then the capacity can be less
than 2Mbps, and higher than 1.1Mbps
Generally, let where x_i is the
maximum rate, and \alpha_i \in (0,1) is called
effective bandwidth
Definition Overbooking means admitting more
demand than the system can actually support,
knowing that most likely they wont transmit
at the maximum rate at the same time. Overbook is
also used in transportation and hotels to improve
efficiency of resource usage.
12
Communication service- Development of the market

• Internet is a stupid network
• - push the complexity to the edge
• - major reason of its success
• - benefit local telecom companies, but weaken
  the companies at the Internet backbone
• Paradox of the best network
• - the best network is the hardest one to make
  money running (Isenberg and Weinberger, 2001)
• - backbone (long-haul) service has been
  commoditized
• - however, competition in the backbone is the
  severest
• Dot-com bubble
• - overestimated and overinvested in the
  long-haul bandwidth
• - only about 5 of the optical fibre are lit.
• Future
• - hard to say
• - will long-haul bandwidth become scarce again?

13
Communication service- the role of economics

• Theory from economics provides decentralized
  control mechanisms
• - capture the selfish and decentralized behavior
  of network entities
• - guarantee the distributed control move towards
  equilibrium
• - use resource more efficiently
• Network design from a holistic view
• - traditionally only optimize the system
  performance
• - now adopt the economic performance,
  including
• a) system performance
• b) flexibility in usage
• c) ability to adapt and customize the service

14
Outline

• Communication service
• Generic Questions
• - Overprovision or Control?
• - Using price for control and signaling
• Tariffs and Contracts
• Pricing a single link

15
Generic questions- Overprovision or Control?

• To guarantee certain performance level, one can
• 1) overprovision a capacity that is way larger
  than demand
• 2) regulate the demand, i.e., congestion
  control, admission control, etc.
• Advantage of the second approach
• - no need to accurately predict the demand
• - it is hard to overdesign the whole network
• - reservation can be costly
• - provide quality differentiation
• - too much capacity leads to
• commodity

There is a 35 minutes wait. However if youre
willing to sit in hell, I will get you a table
immediately.
16
Generic questions- Use price for control and
signaling

• Price can reduce congestion, and increase
  stability
• - when demand increases, the capacity should
  accordingly expand to guarantee a fixed
  congestion level
• - however, this can not be done in real-time
• - during the transient phase, price can serve
  an important role in increasing stability if we
  charge more with higher congestion level
• Incorporate pricing mechanism into TCP?
• - avoid TCP cheating, provide incentive-compatib
  ility
• Communication between network operation and the
  end users
• - predict traffic from users feedback on the
  tariff provided
• - a good tariff design should be
  incentive-compatible, i.e., the user has no
  incentive to cheat about his actual usage plans

In sum, the charged price (for a customer) should
be a function of her sending rate and the
congestion level of the network
17
Outline

• Communication service
• Generic Questions
• Tariffs and Contracts
• Pricing a single link

18
Tariffs and Contracts

• Tariff is a service plan that defines how the
  service is going to be
• charged
• - it affects the behavior
• - a good tariff design should encourage
  good-minded behavior

An example from taxi fee The price of one trip is
calcuated as abTcX (T time, X speed), and
the time and speed are metered exclusively. If
speed is lower than certain value, time
counts if speed is higher than this value, speed
counts. How does this tariff affects the drive
pattern? Is it incentive-compatible?
19
Tariffs and Contracts

• Another example dynamic pricing in an Internet
  Cafe

A user (who wants to enter the Internet Cafe)
pays a fixed price, say 3, and gets an access
time T, which depends on the time and number of
busy terminals (n).
150 minutes during off period (1am-9am) 90
minutes if 0ltnlt150 60 minutes if 150ltnlt300
30 minutes if 300ltnlt450 150 minutes if
0ltnlt150 120 minutes if 150ltnlt300 90 minutes if
300ltnlt450
during peak time (11am-3pm)
T
during normal time (all other time)
Effects of this tariff? - lower price for
off-peak times helps reduce the peak demand - a
feedback system like a thermostat
20
Tariffs and Contracts

• Example of contracts

Important factors - statistical or static - who
chooses h
21
Outline

• Communication service
• Generic Questions
• Tariffs and Contracts
• Pricing a single link

22
Pricing a single link

• Preliminary model (chapter 1.4.4)
• - N customers sharing a single link with total
  capacity C bits/second
• - user i is allocated xi bits/second
• - ui(x) is a concave utility function of user i
• - there exists a price p such that the above
  problem can be solved
• in a decentralized way, by letting each
  user i solve the problem
• - we can solution xi(p), expressed as a
  function of p

23
Pricing a single link

• Preliminary model
• - we define xi(p) as the demand function of user
  i
• - xi(p) strictly decrease with p
• - we can start at a very low value of p such
  that ?xi(p)gtC, and then
• gradually increase p until at some point
  p, we have ?xi(p)C.
• - Now we get the optimal price p, such that the
  capacity is fully utilized, and the total value
  to the society is maximized, meanwhile, each user
  also wants to cooperate by solving problem P1 !

24
Pricing a single link

• Pricing with statistical multiplexing (chapter
  4.1-4.4)
• - Now we consider some overbooking is allowed
• user i has a peak sending rate xi, and an
  effective bandwidth ai, the
• capacity constraint is now changed to
  ?aixi C
• - By Lagrangian method, one can show that the
  maximum capacity constraint is active if only
  there exists a scalar ? such that ui(xi) ?ai
• - Let pi ?ai, then the problem can be solved by
  letting each user solve
• - Observe that the effective bandwidth is
  proportional to pi, this illustrates the
  motivation of using effective bandwidth in
  pricing.
• - Question remains how to find effective
  bandwidth for different network technologies and
  applications?