Outlinesmall Rate control - stable, efficient and fair share of networkssmall Rate control - stable, efficient and fair share of networkssmall Rate control - stable, efficient and fair share of networkssmall Rate control - stable, efficient and fair share of networksRate control : avoid congestion collapseOutlineTCP -- Transport Control ProtocolTCP -- Transport Control ProtocolTCP -- Transport Control ProtocolTCP -- Transport Control ProtocolTCP: sawtooth behaviorTCP: steady state average sending rateTCP: Great success but limited to dataTCP: Great success but limited to dataTCP: Great success but limited to dataOutlineTFRC - TCP-Friendly Rate ControlTFRC - TCP-Friendly Rate ControlTFRC - TCP-Friendly Rate ControlTFRC - Goal: smooth rate controlhspace *{-10mm} TFRC - match TCP average sending ratehspace *{-10mm} TFRC - match TCP average sending rateTFRC vs TCP: UCL->AICIRTFRC vs TCP: Hongkong->BerkeleySeveral observationsSeveral observationsSeveral observationsOutlinehspace *{-10mm}Packet loss may caused by channel errorhspace *{-10mm}Packet loss may caused by channel errorhspace *{-10mm}Packet loss may caused by channel errorDesign new protocols for wireless?TCP over wireless: Tons of effortsTCP over wireless: Tons of effortsExplicit Congestion Notification (ECN)Indirect TCP: split the connectionLink-layer: local retransmissionPerformance Characterization ICross layer information integrationEnd-to-end statistics basedSeveral questions unansweredOutlineMotivationMotivationMotivationMotivationsmall Goal: highest throughput, lowest end-to-end packet loss ratesmall Goal: highest throughput, lowest end-to-end packet loss rateed Conditional for Underutilizationed Conditional for Underutilizationed Conditional for Underutilizationed Conditional for UnderutilizationExisting Worked Questionlarge Proposal : open multiple TFRC connectionslarge Proposal : open multiple TFRC connectionslarge Proposal : open multiple TFRC connectionslarge Proposal : open multiple TFRC connectionshspace {-2mm}ormalsize Optimal Number of Connections Exists for Given Settingshspace {-2mm}ormalsize Optimal Number of Connections Exists for Given Settingshspace {-2mm}ormalsize Optimal Number of Connections Exists for Given Settingshspace {-2mm}ormalsize Optimal Number of Connections Exists for Given Settingshspace {-2mm}ormalsize Optimal Number of Connections Exists for Given SettingsExperimental EvidenceHow to Implement - MULTFRCHow to Implement - MULTFRCHow to Implement - MULTFRCHow to Implement - MULTFRCHow to Implement - MULTFRCNS-2 Simulation Settingsormalsize MULTFRC: Performance CharacterizationAdaptivity to Channel ConditionMULTFRC Does not Starve TCPMULTFRC: Works in PracticeSeveral ObservationsOutlineTowards the best approachRate Control for Streaming Video –from Wired to WirelessMinghua [email protected], Spring 04’Video and Image Processing Lab, Department of EECSUniversity of California at Berkeleyhttp://www-video.eecs.berkeley.eduNote: parts of the figures are adapted from slides from I. Stoica, S. Floyd, H. Balakrishnanand W. Tan.M. Chen, EE290T : rate control for streaming video – from wired to wireless – p. 1/45OutlineBackground on rate controlRate control for data over wired networkRate control for streaming video over wired networkRate control for data over wirelessRate control for streaming video over wirelessConclusions and discussionsM. Chen, EE290T : rate control for streaming video – from wired to wireless – p. 2/45Rate control - stable, efficient and fair share of networksWhat is rate control (or congestion control)End-host changes its sending rate based onfeedback from network (e.g. packet loss)Goal: achieve "title", in a distributed mannerHow to do rate control?negative feedback, e.g. packet loss, decreasesending rateotherwise increase it.Why rate control at all?Bandwidth is shared. If not, serious congestion collapseIn October of ’86, first "congestion collapse":throughput from LBL to Berkeley dropped from32Kbps to 40bps,factor-of-thousand drop!M. Chen, EE290T : rate control for streaming video – from wired to wireless – p. 3/45Rate control - stable, efficient and fair share of networksWhat is rate control (or congestion control)End-host changes its sending rate based onfeedback from network (e.g. packet loss)Goal: achieve "title", in a distributed mannerHow to do rate control?negative feedback, e.g. packet loss, decreasesending rateotherwise increase it.Why rate control at all?Bandwidth is shared. If not, serious congestion collapseIn October of ’86, first "congestion collapse":throughput from LBL to Berkeley dropped from32Kbps to 40bps,factor-of-thousand drop!M. Chen, EE290T : rate control for streaming video – from wired to wireless – p. 3/45Rate control - stable, efficient and fair share of networksWhat is rate control (or congestion control)End-host changes its sending rate based onfeedback from network (e.g. packet loss)Goal: achieve "title", in a distributed mannerHow to do rate control?negative feedback, e.g. packet loss, decreasesending rateotherwise increase it.Why rate control at all?Bandwidth is shared. If not, serious congestion collapseIn October of ’86, first "congestion collapse":throughput from LBL to Berkeley dropped from32Kbps to 40bps,factor-of-thousand drop!M. Chen, EE290T : rate control for streaming video – from wired to wireless – p. 3/45Rate control - stable, efficient and fair share of networksWhat is rate control (or congestion control)End-host changes its sending rate based onfeedback from network (e.g. packet loss)Goal: achieve "title", in a distributed mannerHow to do rate control?negative feedback, e.g. packet loss, decreasesending rateotherwise increase it.Why rate control at all?Bandwidth is shared. If not, serious congestion collapseIn October of ’86, first "congestion collapse":throughput from LBL to Berkeley dropped from32Kbps to 40bps,factor-of-thousand drop!M. Chen, EE290T : rate control for streaming video – from wired to wireless – p. 3/45Rate control : avoid congestion collapseM. Chen, EE290T : rate control for streaming video – from wired to wireless – p. 4/45OutlineBackground on rate controlRate control for data over wired networkRate control for streaming video over wired networkRate control for data over wirelessRate control for streaming video over wirelessConclusions and discussionsM. Chen, EE290T : rate control for streaming video – from wired to wireless – p. 5/45TCP –
View Full Document