EECS228a – Lecture 2Research TopicsJean Walrandwww.eecs.berkeley.edu/~wlrWalrand EECS 228a 52OutlineEconomics of NetworksRoutingCongestion ControlTraffic ModelsWalrand EECS 228a 53Economics of NetworksOutlineHangoverPricing of ServicesCompetition of UsersCompetition of ProvidersSuggested Readings:n http://www.bgsu.edu/departments/tcom/annota.htmn http://info.isoc.org/internet-history/n http://www.spp.umich.edu/ipps/papers/info-nets/Economic_FAQs/FAQs/FAQs.htmlWalrand EECS 228a 54Economics of NetworksHangoverBubble: WiredWalrand EECS 228a 55Economics of NetworksHangoverBubble: WirelessWalrand EECS 228a 56Economics of NetworksHangoverOver-Investmentn Based on unrealistic growth forecastn Overcapacity: Fiber 5x100 in three yearsn Too many companies competing for same marketDebtn Wireless: Expensive spectrum licensesn Fibersn IT in companies: PCs, Servers, NetworksWalrand EECS 228a 57EconomicsKey IdeasValue of services to users: externality, QoS, CoSMarket segmentationFlat rate pricing; congestion pricing; Paris metro pricing; time-of-day pricingIncentive compatibilityInter-ISP settlements; Peering agreementsInternet as a public goodWalrand EECS 228a 58EconomicsValue of ServicesExternality: KazaaValue per bit: email vs. fax vs. pictureValue of bit rate: video stream vs. radioValue of low latency: video stream vs. video conferenceValue of low response time: browsing with DSL vs. browsing with 56kQoS affects value and usageValue of QoS depends on application and userWalrand EECS 228a 59EconomicsMarket SegmentationBusinesses vs. Residential CustomersNetwork Application Providers vs. public Web SitesPrinciple: Charge more users with higher utilityWalrand EECS 228a 60EconomicsDifferentiated PricingExamples: n First Class & Economy in plane: More space but much more expensiven Paris Metro: More expensive Fewer Users Better Service (e.g., Stanford vs. Berkeley?)Suggests Class of Service: n Better service by mechanism: e.g., priority n Better service by fewer users: e.g., expensive network; congestion pricing (e.g., packet marking); time-of-dayAlternative: QoS: You know what you pay forn Service Level Agreement (implementation?)n QoS of accepted calls: end-to-end testWalrand EECS 228a 61EconomicsIncentive CompatibilityHow to discover the user’s willingness to pay?Examples:n California Electricity: Providers offer bids and CA buys cheaper first prices escaladen Highest bidder auction: Spectrum auctionsn Highest gets but two highest payn Second highest price: Incentive compatibleWalrand EECS 228a 62EconomicsCompetitionBasic supply and demand:n More capacity than traffic prices drop and providers go bankruptInternet traffic doubles every year instead of every 100 days …. Quality service is still rare and valuable: n Businesses use video conference over ISDN n Expensive commutes and business traveln Users pay a lot for CATV and pay-per-viewn T1 service expensive: demand existsWalrand EECS 228a 63EconomicsGame TheoryFramework to analyze result of interaction of self-interested agentsSuggests strategies for n Pricing servicesn Peering agreementsn Routingn QoS definitionsn Evolution of industry (e.g., consolidation vs. specialization)Two parts: Games & Mechanism DesignWalrand EECS 228a 64RoutingOutlineMotivationGranularityTypesIssuesWalrand EECS 228a 65RoutingMotivationReduce delays: Avoid OAK NY SFImprove reliability: ProtectionSensor networks: Many open questionsAd Hoc networks: More robust, provide QoSIP/Optical: Improve coordinationWalrand EECS 228a 66RoutingGranularityLight Path: WDMCross-Connect: SONETCircuit: TelephoneLabel Switched Path: MPLS; ATMConnectionPacketWalrand EECS 228a 67RoutingGranularity (cont)Benefit of LSP vs SONET is not obvious:n Consider traffic from SF to NY; If that traffic is essentially constant, then SONET is good enough. If not, LSP/SONET is preferable.n If traffic is self-similar, then fluctuations persist at high rateWalrand EECS 228a 68RoutingTypesOn-line vs. Off-lineCentralized vs. DistributedLink State; Distance Vector; Path VectorSource-based vs. Destination-basedQoS routingAd Hoc; Location-BasedAnt-routing (reinforcement)Unicast vs. multicastProtection routingPeer-to-peer vs. overlayWalrand EECS 228a 69RoutingIssuesBenefitsImplementability:n Scalability: communications required; complexity; convergence timen Robustness: sensitivity to errorsWalrand EECS 228a 70Congestion ControlOutlineMotivationExamplesIssuesWalrand EECS 228a 71Congestion ControlMotivationAt user level: Issues with QoSAt network level: Losses, inefficiency, unfairnessAt switch level: Scalability problemsWalrand EECS 228a 72Congestion ControlExamplesTCPCongestion in routersCall Admission ControlWalrand EECS 228a 73Congestion ControlIssuesFairness vs. OptimalitySimplicityRobustnessWalrand EECS 228a 74Traffic ModelsOutlineWhy bother?TransactionsPacket flowsWalrand EECS 228a 75Traffic ModelsWhy Bother?Network should be robust; not based on detailed traffic assumptionsTraffic characteristics impactn Effectiveness of multiplexingn Buffer sizes requiredn Time scale of bandwidth allocationsWalrand EECS 228a 76Traffic ModelsTransactionsFile transfers: n File sizes: Heavy tailedn Timing of requests: Poissonn Geography: w Kazaa – poor localityw Akamai – improved localityOther applications: n video conferencesn VoIPWalrand EECS 228a 77Traffic ModelsPacket FlowsSelf-Similarity: n Heavy Tail + TCP Self Similar Flowsn Heavy Tail Files + Structure of Web Sites Self SimilarityRelevance: n Not obvious – a matter of time
View Full Document
Unlocking...