Internet Infrastructure and Pricing

1
Internet Infrastructure and Pricing
2
Internet Pipelines

• Technology of the internet enables ecommerce
• Issues of congestion and peak-load pricing
• Convergence of technology and elimination of
  boundaries between different communications
  services
• Telephone
• Cable
• Satellite
• Wireless
• ISPs
• We will examine pricing policies for this
  infrastructure and associated issues of economic
  efficiency

3
Internet Network Architecture
Local Area Networks
End Users
Regional Networks
Backbone Networks
Other Local Networks
Other Regional Networks
4
Congestion

• Network Economics
• Large networks are expensive to build, but cheap
  to operate
• Since fixed costs are sunk costs, network
  operators have incentives to connect as many
  users (at very low marginal cost) to the network
• As the number of users increases, peak use
  congestion will occur
• World Wide Wait
• Hourglass fatigue
• The system then rations the network resource
  (bandwidth) according to user patience rather
  than social value

5
Congestion

• Current efforts to fix the problem
• Engineering fixes
• Expand capacity
• Doomed to failure Growth in demand will keep
  pace with growth in capacity for the foreseeable
  future
• Added problem of misallocation of investment
  resources in response to inefficient use of
  network resources
• Priority schemes
• IPv6 will include priority settings to move the
  network away from the current best-effort,
  first-come-first-served approach to managing
  traffic
• Not incentive compatible, however users can
  modify the settings to boost their own priority

6
Congestion

• Digital tragedy of the commons
• Congestion imposes external costs on internet
  users, leading to misallocation of resources
• Public good aspect of the network
• Large fixed cost
• Zero marginal cost of adding additional packets
  during slack periods
• Positive social costs of adding packets during
  congested periods
• Highway analogy

7
Pricing

• Dynamic optimal pricing (Gupta, Stahl and
  Whinston)
• Studies based on simulations show the feasibility
  of computing accurate, real-time estimates of
  social costs of congestion
• These cost estimates are then combined with
  information about traffic flow through a site,
  packet sizes, and priority class of the site to
  derive optimal prices (marginal social cost of
  processing an additional packet)
• Users establish a preference profile based on a
  menu of options including
• Monetary costs of using the site during peak
  periods for the specified option
• Expected waiting times during peak periods for
  the option

8
Pricing

• Smart agents can then automate the users
  decision process based on the established
  preference profile, using information updates
  provide by the users ISP
• Billing is then based on a standard model of
  wholesale billing practices
• User receives bill from ISP for all internet
  transport services
• ISP receives a bill from regional network
  provider and passes individual user costs on to
  them
• Benefits in simulations of various pricing
  policies

9
Pricing
10
Pricing

• Static priority pricing
• Service discipline Mechanism to allow network
  operators to assign jobs to specific service
  classes
• Different service classes are priced differently
  according to user values for each type of service
  and average resulting congestion on the network
• Scheme maximizes time-averaged user benefits
• Problems with implementation
• Requires that operator be able to collect
  proprietary information about user preferences
  and valuations of each type of service
• Users have incentives to misrepresent valuations
• Static implementation means high-priority users
  will pay too little during peak periods

11
Pricing

• Smart Markets
• Real-time auctions by packets traversing the
  network for router or other network services
• Encode bids in packet headers
• Run uniform auctions over a pre-specified time
  interval, with the n highest bidding packets
  being processed in the interval, and charged the
  n1st highest price.
• If all packets process in the allotted time, the
  price is zero.

12
Pricing

• Problems with the scheme
• Deadweight time loss while packets are sorted
  according to bids
• Efficiency properties of the auction mechanism
  require
• All potential users are present at the auction
• The value of a job is not contingent on what
  happens at other market nodes
• These properties are violated in dynamic networks
• Market operates over time intervals, so packets
  arriving late (even nanoseconds late) can have no
  influence on the price, and hence on the
  allocation of the scarce resource
• In real internet operations, the value of having
  a job processed depends on having all of the
  relevant packets processed, so that in principle,
  prices should depend on valuations attached to
  other packets in a job, and hence, on prices at
  other smart market nodes in the network

13
Pricing

• Connection only and flat rate pricing
• Dominant form of pricing currently
• Connection only fees may vary according to
  bandwidth option selected, but make no attempt to
  price for load
• Some online services charge per unit of time
  connected, though these tend to be for dial-up
  modem connections, as opposed to always-on
  connection (DSL, cable)
• Flat rate pricing charges a single rate for
  connection regardless of time used
• AOLs early experience
• Serious congestion effects when prices are too
  low
• Possibility of using flat rates to implement
  averaged optimal prices

14
Pricing

• Policy issues
• Imposition of optimal pricing requires some
  centralized regulation of the network in order to
  internalize the costs of congestion
• Current climate is one of privatization
• Likely outcome
• Excess investment in engineering solutions
  expanded bandwidth distributed uniformly
  throughout the network
• Continued congestion effects as demand for
  bandwidth increases
• Possibility for rational pricing on intranets