Price Discovery at Network Edges

1
Price Discovery at Network Edges

• G. S. Arora, M. Yuksel, S. Kalyanaraman,
• T. Ravichandran and A. Gupta
• Rensselaer Polytechnic Institute, Troy, NY

2
Overview

• Motivation
• Edge-to-Edge Concepts
• Price Discovery Framework
• Pricing Schemes
• Simulations
• Summary

3
Motivation

• Need for new economic models
• Adaptive, particularly congestion, pricing is
  necessary
• Implementation problems need to be solved
• Upgrades should be limited
• Administrative access is necessary
• Can we do it only at network edges?

4
Edge-to-Edge Concepts

• A generic view and the trend for the Internet.
• Administrative access is available at edges.
• So, possible to coordinate ingress and egress
  edge stations.
• A more complex version of Clarks Edge Pricing is
  possible..

5
Price Discovery Framework

• Given edge-to-edge coordination, estimate
  edge-to-edge capacity .
• Congestion pricing ? Pricing of edge queue
  .
• Severity of congestion is the ratio .
• Is this ratio really a good parameter?
• Formulate objective function
• Regression analysis for unknown variables of the
  objective function

6
Price Discovery Framework (contd)

• Let there be k observations in a contract period.
• User adaptation
• Actual spent budget for user with reservation
  price

7
Price Discovery Framework (contd)

• Objective Minimize non-utilized capacity while
  keeping edge queue less than a pre-defined value
  .
• Formulation
• subject to

8
Price Discovery Framework (contd)

• Other than Bk everything else is known.
• For k1000, qmax50, Ck N(98,2) truncated in
  the range 96,100
• Regression analysis for Bk U(20,50) and Bk
  U(30,150) verified that qi/Ci,mean is strongly
  associated with optimal price pi.
• So, we can use the predictor qi/Ci,mean to
  determine pi in adaptive pricing at the edge.

9
Pricing Schemes

• Assuming that ISP wants to keep edge queue in the
  range ql, qh.
• Proportional Increase Proportional Decrease
  (PIPD)
• Proportional Increase Additive Decrease (PIAD)

10
Pricing Schemes (contd)

• Additive Increase Additive Decrease (AIAD)
• Additive Increase Proportional Decrease (AIPD)

11
Simulations

• Define user demand according to reservation price
  and
• Initial parameters
• Step increase in demand In times (50, 100), i.e.

12
Simulations (contd)
13
Simulations (contd)
14
Simulations (contd)

• PIPD and PIAD performs significantly better than
  AIPD and AIAD.
• Compared to PIPD, PIAD has less variation in
  price but utilization is slightly less too.
• So, the best one is either PIAD or PIPD,
  depending on value of utilization.

15
Simulations (contd)

• Investigated effect of several parameters on PIAD
  performance.
• We run two users with different and
  reservation prices

16
Simulations (contd)
17
Summary

• Adaptive (particularly congestion) pricing is
  necessary for enabling better economic models.
• Price Discovery
• Deployable over diff-serv
• Possible to implement congestion pricing at edges
• Possible to develop variety of pricing schemes
  PIPD, PIAD, AIPD, AIAD.