Outline EECS228a Lecture 2 Economics of Networks Research Topics - PDF document

Walrand Lecture 2 Outline EECS228a – Lecture 2 Economics of Networks Research Topics Routing Congestion Control Jean Walrand Traffic Models www.eecs.berkeley.edu/~wlr Walrand EECS 228a 52 Economics of Networks Economics of Networks Outline Hangover Bubble: Wired Hangover Pricing of Services Competition of Users Competition of Providers Suggested Readings: n http://www.bgsu.edu/departments/tcom/annota.htm n http://info.isoc.org/internet-history/ n http://www.spp.umich.edu/ipps/papers/info- nets/Economic_FAQs/FAQs/FAQs.html Walrand EECS 228a 53 Walrand EECS 228a 54 Economics of Networks Economics of Networks Hangover Hangover Bubble: Wireless Over-Investment n Based on unrealistic growth forecast n Overcapacity: Fiber 5x100 in three years n Too many companies competing for same market Debt n Wireless: Expensive spectrum licenses n Fibers n IT in companies: PCs, Servers, Networks Walrand EECS 228a 55 Walrand EECS 228a 56 EECS 228a 1

Walrand Lecture 2 Economics Economics Key Ideas Value of Services Externality: Kazaa Value of services to users: externality, QoS, CoS Value per bit: email vs. fax vs. picture Market segmentation Value of bit rate: video stream vs. radio Value of low latency: video stream vs. video Flat rate pricing; congestion pricing; Paris conference metro pricing; time-of-day pricing Value of low response time: browsing with Incentive compatibility DSL vs. browsing with 56k Inter-ISP settlements; Peering agreements QoS affects value and usage Internet as a public good Value of QoS depends on application and user Walrand EECS 228a 57 Walrand EECS 228a 58 Economics Economics Market Segmentation Differentiated Pricing Examples: Businesses vs. Residential Customers n First Class & Economy in plane: More space but much more expensive Network Application Providers vs. public n Paris Metro: More expensive Fewer Users Better Service (e.g., Stanford vs. Berkeley?) Web Sites Suggests Class of Service: Principle: Charge more users with n Better service by mechanism: e.g., priority n Better service by fewer users: e.g., expensive network; higher utility congestion pricing (e.g., packet marking); time-of-day Alternative: QoS: You know what you pay for n Service Level Agreement (implementation?) n QoS of accepted calls: end-to-end test Walrand EECS 228a 59 Walrand EECS 228a 60 Economics Economics Incentive Compatibility Competition Basic supply and demand: How to discover the user’s willingness to pay? n More capacity than traffic prices drop and Examples: providers go bankrupt n California Electricity: Providers offer bids and CA Internet traffic doubles every year instead of buys cheaper first prices escalade every 100 days …. n Highest bidder auction: Spectrum auctions Quality service is still rare and valuable: n Highest gets but two highest pay n Businesses use video conference over ISDN n Second highest price: Incentive compatible n Expensive commutes and business travel n Users pay a lot for CATV and pay-per-view n T1 service expensive: demand exists Walrand EECS 228a 61 Walrand EECS 228a 62 EECS 228a 2

Walrand Lecture 2 Economics Routing Game Theory Outline Framework to analyze result of interaction of Motivation self-interested agents Granularity Suggests strategies for Types n Pricing services n Peering agreements Issues n Routing n QoS definitions n Evolution of industry (e.g., consolidation vs. specialization) Two parts: Games & Mechanism Design Walrand EECS 228a 63 Walrand EECS 228a 64 Routing Routing Motivation Granularity Reduce delays: Avoid OAK NY SF Light Path: WDM Improve reliability: Protection Cross-Connect: SONET Sensor networks: Many open questions Circuit: Telephone Ad Hoc networks: More robust, provide Label Switched Path: MPLS; ATM QoS Connection IP/Optical: Improve coordination Packet Walrand EECS 228a 65 Walrand EECS 228a 66 Routing Routing Granularity (cont) Types On-line vs. Off-line Benefit of LSP vs SONET is not obvious: Centralized vs. Distributed n Consider traffic from SF to NY; If that Link State; Distance Vector; Path Vector traffic is essentially constant, then SONET Source-based vs. Destination-based is good enough. If not, LSP/SONET is QoS routing preferable. Ad Hoc; Location-Based n If traffic is self-similar, then fluctuations Ant-routing (reinforcement) persist at high rate Unicast vs. multicast Protection routing Peer-to-peer vs. overlay Walrand EECS 228a 67 Walrand EECS 228a 68 EECS 228a 3

Walrand Lecture 2 Routing Congestion Control Issues Outline Benefits Motivation Implementability: Examples n Scalability: communications required; Issues complexity; convergence time n Robustness: sensitivity to errors Walrand EECS 228a 69 Walrand EECS 228a 70 Congestion Control Congestion Control Motivation Examples At user level: Issues with QoS TCP At network level: Losses, inefficiency, Congestion in routers unfairness Call Admission Control At switch level: Scalability problems Walrand EECS 228a 71 Walrand EECS 228a 72 Congestion Control Traffic Models Issues Outline Fairness vs. Optimality Why bother? Simplicity Transactions Robustness Packet flows Walrand EECS 228a 73 Walrand EECS 228a 74 EECS 228a 4

Walrand Lecture 2 Traffic Models Traffic Models Why Bother? Transactions File transfers: Network should be robust; not based on n File sizes: Heavy tailed detailed traffic assumptions n Timing of requests: Poisson Traffic characteristics impact n Geography: w Kazaa – poor locality n Effectiveness of multiplexing w Akamai – improved locality n Buffer sizes required Other applications: n Time scale of bandwidth allocations n video conferences n VoIP Walrand EECS 228a 75 Walrand EECS 228a 76 Traffic Models Packet Flows Self-Similarity: n Heavy Tail + TCP Self Similar Flows n Heavy Tail Files + Structure of Web Sites Self Similarity Relevance: n Not obvious – a matter of time scale Walrand EECS 228a 77 EECS 228a 5