A Comparative Evaluation of Internet Pricing Schemes: Smart Market and Dynamic Capacity Contracting

1
A Comparative Evaluation of Internet Pricing
Schemes Smart Market and Dynamic Capacity
Contracting

• T. Ravichandran
• Shiv Kalyanaraman
• Ranjita Singh
• Murat Yuksel,
• Rensselaer Polytechnic Institute

2
Overview

• Background of this research
• Motivations and research objectives
• Internet pricing models
• Differentiated Services Architecture
• Dynamic Capacity Contracting (DCC)
• Implementation of DCC and Smart Market
• Experimental results
• Findings
• Future Directions

3
Background

• Started with work in congestion control and
  traffic management
• Need to integrate economic and technical measures
  to effectively manage congestion
• Funded by a 3-year NSF grant
• Interdisciplinary work
• Proof of concept stage

4
Motivations

• Exponential growth in internet traffic
• Technical solutions such as better traffic
  management and over provisioning have been used
• Economic measures such as responsive pricing have
  been proposed but not been implemented
• Implementation issues have not been addressed in
  most pricing models
• Need for better than best effort service ability
  to provide service discrimination

5
Objectives

• To develop internet pricing schemes taking into
  consideration
• the implementation constraints
• the service discrimination features of emerging
  internet architectures such as the Diff-Serv
• transaction overhead and/or accounting costs
• To evaluate the pricing schemes for both
  technical and economic efficiencies

6
Internet Pricing Models

• Static Pricing Models
• Flat rate pricing
• Expected capacity contracting (Clark, 1999)
• Dynamic Pricing Models
• Usage-based pricing
• Congestion-sensitive pricing
• Smart Market (MacKie- Mason Varian, 1995)
• Priority-class pricing (Gupta, Stahl Whinston,
  1999)

7
Differentiated-Services(Diff-Serv) Model
End system
End system
Interior Router
Egress Edge Router
Ingress Edge Router

• A standard architecture for the Internet
• complex operations at network edges (i.e. edge
  routers)
• simple operations in network core (i.e. interior
  routers)
• Expected to be the choice of ISPs and bandwidth
  providers
• Protocols for Service Level Agreement (SLA) are
  already available
• Possible to make congestion-based pricing at the
  edges

8
Dynamic Capacity Contracting

• Extends Clarks expected capacity contracting
  model to incorporate short-term contracts and
  adds mechanisms to make it congestion-sensitive
• Customers enter into short-term contract with the
  service provider.
• Contract is specified as follows
• contract for a given traffic class is a function
  of volume (number of bytes), contract term (time
  units) and price per unit volume
• Contracted volume handled with a low probability
  of delay and packet loss
• Volume in excess of the contracted volume handled
  with best-effort service
• Customers charged only as per contract
  irrespective of the actual volumes sent

9
Dynamic Capacity Contracting

• Short-term contracts are used to provide needed
  flexibility to change the price per unit volume
  based upon congestion
• Price calculation
• price is set by matching demand and supply
• network capacity adjusted to reflect the current
  congestion levels
• price Aggregate demand/Adjusted Capacity
• P SBi / min(average_rate_limit,
  bottleneck_capacity)T where SBi is the total
  contracted amount for the previous contract term.
• The average_rate_limit is updated each contract
  term and is based upon network congestion.

10
Smart Market

• A congestion-sensitive pricing scheme proposed by
  MacKie-Mason Varian (1993).
• Imposes a per-packet-charge that reflects
  marginal congestion costs.
• A Vickery auction model for price determination
• Scenario
• Users assign a bid value for each packet and
  the packet tries to make through the network.
• Each packet has a probability of being dropped
  depending on the current threshold (cutoff) bid
  value among the routers in the network, which
  depends on congestion level at the particular
  router, and is adjusted by that router.
• Finally, users pay the highest threshold value
  that the packet passed through, which is the
  market-clearing price.

11
Comparison of the Models
12
Implementation of the Pricing Models

• We use a simple network configuration in our
  simulation experiments
• single bottleneck with a rate of 1Mbps
• customers send constant bit rate UDP traffic with
  fixed packet sizes (1000 bytes)
• contract term in DCC and the length of the
  feedback time interval in the smart market are
  set to be 0.4sec
• the length of the observation interval in DCC is
  set to 80ms

13
DCC Implementation

• Flowchart of actions for the customer

14
DCC Implementation

• Flowchart of actions for the provider (Ingress
  edge router)

15
DCC Implementation

• Updating average_rate_limit requires congestion
  indication from the egress edge.
• How to update average_rate_limit
• Sub-divide the terms into smaller observation
  intervals
• Define rate_limit for each observation interval
• average_rate_limit mean of each of the
  rate_limits in the observation intervals

16
Smart Market Implementation

• Smart market implementation issues
• packet-ordering as per bid values conflicts with
  TCP packet ordering
• assumes instant feedback of clearing price to
  customers which is not feasible in a real wide
  area network
• We use deterministic time intervals at the
  routers (edge and interior) as a way to handle
  the feedback problem. Customers get feedback from
  the network at the end of each time interval
  thereby they can make adjustments to their bid
  values and demands.
• The length of this time interval is a comparable
  measure to the length of contract term in DCC.
  This makes the two schemes comparable

17
Smart Market Implementation

• Flowchart of actions of the customer

18
Smart Market Implementation

• Flowchart of actions for the provider (ingress
  edge router)

19
Smart Market Implementation

• Flowchart of actions for the provider (egress
  edge router)

20
Comparative Evaluation

• Technical Efficiency
• Bottleneck utilization
• Throughput and goodput
• Queue length
• Economic efficiency
• Fairness in volume allocation

21
Results

• Total volume allocated to all customers is
  significantly higher in the case of DCC. This
  indicates that DCC better utilizes the
  bottleneck.

22
(No Transcript)
23
Results

• DCC achieves higher throughput and goodput with a
  lower packet drop.

24
Results

• Both models seem to do well in clearing queues at
  the bottleneck

25
Results

• Smart market allocates volumes more fairly than
  DCC

26
Results
27
Results
28
Summary
29
Future Directions
30
Questions and Comments