nossdav99pk

Service Differentiation: Congestion Pricing, Brokers and Bandwidth Futures: 

Peter Key peterkey@microsoft.com http://research.microsoft.com/network/disgame.htm Service Differentiation: Congestion Pricing, Brokers and Bandwidth Futures

Sine qua non ...: 

Sine qua non ... Differential Services / QOS  Differential pricing!

Outline: 

Outline Background Congestion Pricing A game …. Commodity markets and Futures

Access to / Control of Scarce Resources: 

Access to / Control of Scarce Resources Airline seats, electrical power, economic wealth Are networks any different? Control technologies not independent or implementation /economics ATM: eg sensible relative tariffs, CBR 1, VBR 2, ABR .1, UBR .01, … result?

Resource system (‘network’) : 

Resource system (‘network’) Resource j  Capacity Cj User /route r  Ajr links users to resources

Freebyte: 

Freebyte The network is free most of the time! Only pay when busy (incremental cost, small) … aka Congestion Pricing Send a signal to users when traffic that should not be carried enters (moveable threshold)

Basic Idea: 

Basic Idea Users generate load (packets) Network sends back signals (load dependent) Signals : proportional to load Act as feedback indicators Represent pricing signals marginal incremental costs (derivatives …) congestion costs real money or virtual / distributed mint

Matching Prices to Load: 

Matching Prices to Load Theorem: For linear increase multiplicative feedback flow control family, cannot have demand =capacity if prices are bounded for fixed capacity and arbitrary number number of flows. If prices are unbounded (bad!) then can have demand  capacity as number of flows increase, Else: Match willingness to pay to price capacity

Sample Path Shadow Prices: 

Sample Path Shadow Prices

Shadow path shadow prices -buffered model: 

Shadow path shadow prices -buffered model B B

Optimisation Framework (for fairness): 

Optimisation Framework (for fairness) System optimum User optimum C is cost function,eg Consistent set of taxes (prices) and load exist s.t. user opt = social opt.

Example - elastic control: 

Example - elastic control eg

Applications: 

Applications Virtual Private Networks Non-TCP apps Web Media MBONE Tools, RealMedia, Netshow Real application uses (like a DB)

Aggregates: 

Aggregates All the channels in a VPN connection Audio/Video/Data in a conference “Working set” for web server Policy based routers/firewalls

Aggregates: Audio / Video: 

Aggregates: Audio / Video Policy scheduler Network User/ end system

Network vs Users: 

Network vs Users “My work is a game, a very serious game” Escher Users Signals Data/Info Network

Flow Control and Marking Strategies: 

Flow Control and Marking Strategies Definition: The phenotype of an individual plant or animal … or flow-control scheme … is a consequence of the interaction between its genotype and the environment in which it exists

Distributed Multi-player Game: 

Distributed Multi-player Game Internet MSR Cambridge

Example Game: 

Example Game Transfer a given amount of data F at minimum cost in time T Strategies: CBR, if stationary regime, optimal if price “convex” in region (lightly loaded) else, have to predict fluctuations

WTP vs TCP: 

WTP vs TCP

The User interface …: 

The User interface … Must be simple!

Guaranteed Flows / Streams?: 

Guaranteed Flows / Streams? Probabilistic guarantees OK, provided proportion of adaptive traffic adequate Just a question of risk - brokers (cf fixed rate mortgages) Futures market: a reservation is a (European) call option for an amount of bandwidth, where fair price = Integral of expected price over expected life-time

Bandwidth Exchanges: 

Bandwidth Exchanges 9 Inteconnection operators (clearing houses, brokers etc) Band-X, London 1997 RateXchanges, anonymous, PSTN voice, VOIP, b/width

Is Bandwidth Commodity?: 

Is Bandwidth Commodity? commodity A physical substance, such as food, grains, and metals, which is interchangeable with other product of the same type, and which investors buy or sell, usually through futures contracts. Or more generally, a product which trades on a commodity exchange Originally, basic, unrefined (wheat not flour)

Commodity Markets: 

Commodity Markets History repeats itself … (Fora Vendalia) Japanese forward contract in 1730 for rice In China 4000BC??? Chicago forwards 1833+, cash markets Standards (quality / quantity evolved …) Options on futures, 1983 Forward contracts aimed at delivery, now only a few % complete

Futures and Auctions: 

Futures and Auctions Market efficient? (Sorros ….) Commodity prices highly random (small trend) Auctions: FCC holds b/width auctions

Futures pricing: 

Futures pricing Assume current rate pt Fair price for European option at time t lasting expected time T given current spot price s=p0, (constant load)

Second Price Auctions: 

Second Price Auctions Suppose users are prepared to pay amount per packet, (eg fixed data, maximum cost), if price goes above this, will drop out In a mixed environment, those left pay this clearing price ie Vickrey auction via congestion pricing

Timescales: 

Timescales Connection Reaction (RTT) Packet Level Seconds line rate ms Application Network ms Critical timescale

Critical timescales : 

Critical timescales

Facilitators: 

Facilitators Small buffers (compared to transmission delay) Target loads below 100% … Simple feedback signal, eg ECN bit/byte Signal reflects costs but … User interface simple (risk apportionment)