160 likes | 165 Views
Pricing in Computer Network: Reshaping the Research Agenda. Authors: S. Shenker, D. Clark, D. Estrin and S. Herzog. Presenter: Lihua Yuan Feb 15, 2004, ECS 289L Network Pricing. The Authors. Scott Shenker: Integrated Service Dave Clark: E2E argument vs. the brave new world
E N D
Pricing in Computer Network:Reshaping the Research Agenda Authors: S. Shenker, D. Clark, D. Estrin and S. Herzog Presenter: Lihua Yuan Feb 15, 2004, ECS 289L Network Pricing
The Authors • Scott Shenker: Integrated Service • Dave Clark: E2E argument vs. the brave new world • Deborah Estrin: Routing Arbitor, VINT • S. Herzog: ? The Blind Men and the Elephant
Outline • Current research – the optimality paradigm • Critiques • Is marginal cost relevant? • Accessible? • Is optimality the only goal? • New pricing paradigm – Edge Pricing • Architecture Issues
Current Research • Flat pricing <> Usage-based pricing • Assume usage-based pricing is necessary • Find “optimal” in a simplified model • Price marginal congestion costs
Critiques 1 – Is marginal congestion cost relevant? • Marginal cost is a function of demand and supply, • marginal cost > facility cost + operation cost ? • Stable competitive equilibrium • Total charge = attachment fee + usage fee • Marginal cost could be different from different sub populations of users • Every subset has its own utility function • Competition aimed at a certain subset? • eg. Mobile phone service plans?
Critiques 2 – Is Marginal Costs Accessible? • Does a simple utility function exist in reality? • E.g. Utility a function of bandwidth or delay, • Anybody doing this in real life? • Different users different requirements • Per-packet based charge vs. Flow-based utility • Users don’t think at packet level • Packet-utility relationship ever changing • Adaptive applications, improved TCP
Critique 3 – Is Optimality the Only Goal? • Optimality in simplified model ignored too many basic structural issues • Pricing should be compatible with structure of network applications • Encourage multicast • Charge receiver or sender? • Compatible with network service market • Multilateral vs. Bilateral
Critique 3 – Is Optimality the Only Goal? • Local control more important than absolute optimal • Optimality single pricing scheme across the network • Locality in pricing • No truly optimal scheme anyway
Edge Pricing – Approximating congestion costs 1. Replace the current congestion condition by expected congestion condition • QoS-sensitive time-of-day pricing • Insensitive to instant fluctuation • Not encourage dynamic adaptation , but time-shift • User monitoring and adaptatio (Mobile phone again ) • Variability at Edge – Difficult for network to access
Edge Pricing – Approximating congestion costs • Replace the cost of the actual path with the cost of the expected path • Depends only on Source-Destination pair • Dependency on routing brings extra dynamics • Easier for user to adapt
Edge Pricing – Local Determination • Prices determined locally at access point (network edge) • Not distributed computed alone the entire path • May depend on information collected from the whole network • Bilateral relationship • Billing structure completely local
Edge Pricing – Forms • Usage-constraining prices • Two extreme cases • Flat rate (up to certain capacity, this assumption often hidden) • Per-packet/per-reservation charge
Architectural Issues – Multicast Approach 1: identify every receiver Approach 2: compute cost on-the-fly
Architectural Issues – Charging Receivers • How does a receiver indicate its willingness to pay? • RSVP • How to bill the receiver? • At the receiver’s access point • Charges are accumulated along the path
Architectural Issues – Open Issues • How to charge? • S/D pair (go Dutch?) • multicast • price depend on # of audience? • Free ride? (ABC)
Questions? • I got one! • Why network engineers need to worry about sender-pay or receiver-pay scheme? Can’t sender bill the receiver later/beforehand?