Problem and Objectives Background NUM with packet corruption Examples Rate Control with Packet Corruption David Hayes david.hayes@ieee.org Joint work the Lachlan Andrew CSIRO ICT Centre, Marsfield, 27 August 2007 David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 1/ 33 Problem and Objectives Background NUM with packet corruption Examples Outline 1 Problem and Objectives 2 Background 3 NUM with packet corruption 4 Examples 5 Conclusions and further work David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 2/ 33 Problem and Objectives Background NUM with packet corruption Examples Problem At what rate should traffic sources send when experiencing packet corruption along their path? David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 3/ 33 Problem and Objectives Background NUM with packet corruption Examples Problem At what rate should traffic sources send when experiencing packet corruption along their path? What do we mean by corruption? Information either completely lost, or partially lost. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 3/ 33 Problem and Objectives Background NUM with packet corruption Examples Problem At what rate should traffic sources send when experiencing packet corruption along their path? What do we mean by corruption? Information either completely lost, or partially lost. Wireless links can have very high corruption rates eg Sensor Networks David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 3/ 33 Problem and Objectives Background NUM with packet corruption Examples Problem At what rate should traffic sources send when experiencing packet corruption along their path? What do we mean by corruption? Information either completely lost, or partially lost. Wireless links can have very high corruption rates eg Sensor Networks Why is it bad? Reduction in end-user utility TCP treats corruption loss as congestion loss David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 3/ 33 Problem and Objectives Background NUM with packet corruption Examples Problem At what rate should traffic sources send when experiencing packet corruption along their path? What do we mean by corruption? Information either completely lost, or partially lost. Wireless links can have very high corruption rates eg Sensor Networks Why is it bad? Reduction in end-user utility TCP treats corruption loss as congestion loss What can be done? FEC — adaptive? Layer 2 correction Transport layer correction David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 3/ 33 Problem and Objectives Background NUM with packet corruption Examples Problem At what rate should traffic sources send when experiencing packet corruption along their path? Examples Corruption close to source David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 3/ 33 Problem and Objectives Background NUM with packet corruption Examples Problem At what rate should traffic sources send when experiencing packet corruption along their path? Examples Corruption close to source David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 3/ 33 Problem and Objectives Background NUM with packet corruption Examples Problem At what rate should traffic sources send when experiencing packet corruption along their path? Examples Corruption close to source David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 3/ 33 Problem and Objectives Background NUM with packet corruption Examples Problem At what rate should traffic sources send when experiencing packet corruption along their path? Objectives Corruption and optimal fluid based congestion control. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 3/ 33 Problem and Objectives Background NUM with packet corruption Examples Problem At what rate should traffic sources send when experiencing packet corruption along their path? Objectives Corruption and optimal fluid based congestion control. What is a fair transmission rate? David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 3/ 33 Problem and Objectives Background NUM with packet corruption Examples Problem At what rate should traffic sources send when experiencing packet corruption along their path? Objectives Corruption and optimal fluid based congestion control. What is a fair transmission rate? Look at different ideas of fairness. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 3/ 33 Problem and Objectives Background NUM with packet corruption Examples Problem At what rate should traffic sources send when experiencing packet corruption along their path? Objectives Corruption and optimal fluid based congestion control. What is a fair transmission rate? Look at different ideas of fairness. Propose this as an objective for TCP modifications. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 3/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Background ICCRG and IETF Internet Congestion Control Research Group (ICCRG) How should TCP be adapted to suit the evolving Internet. TCP’s rate when there is loss due to corruption. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 4/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Background ICCRG and IETF Internet Congestion Control Research Group (ICCRG) How should TCP be adapted to suit the evolving Internet. TCP’s rate when there is loss due to corruption. The Datagram Congestion Control Protocol (DCCP). David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 4/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Background Price based rate control [Kelly97] developed an optimisation model to investigate pricing on broadband networks. Network and User sub-problems David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 5/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Background Price based rate control [Kelly97] developed an optimisation model to investigate pricing on broadband networks. Network and User sub-problems User chooses a willingness to pay ws based on a price. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 5/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Background Price based rate control [Kelly97] developed an optimisation model to investigate pricing on broadband networks. Network and User sub-problems User chooses a willingness to pay ws based on a price. Network chooses source rates xs based on ws David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 5/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Background Price based rate control [Kelly97] developed an optimisation model to investigate pricing on broadband networks. Network and User sub-problems User chooses a willingness to pay ws based on a price. Network chooses source rates xs based on ws Low and Lapsley [Low99] solve the same optimisation problem in a different way: David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 5/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Background Price based rate control [Kelly97] developed an optimisation model to investigate pricing on broadband networks. Network and User sub-problems User chooses a willingness to pay ws based on a price. Network chooses source rates xs based on ws Low and Lapsley [Low99] solve the same optimisation problem in a different way: User choose a send rate xs given path cost qs . David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 5/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Background Price based rate control [Kelly97] developed an optimisation model to investigate pricing on broadband networks. Network and User sub-problems User chooses a willingness to pay ws based on a price. Network chooses source rates xs based on ws Low and Lapsley [Low99] solve the same optimisation problem in a different way: User choose a send rate xs given path cost qs . Network chooses a price for each link pl based on congestion. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 5/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Background Price based rate control [Kelly97] developed an optimisation model to investigate pricing on broadband networks. Network and User sub-problems User chooses a willingness to pay ws based on a price. Network chooses source rates xs based on ws Low and Lapsley [Low99] solve the same optimisation problem in a different way: User choose a send rate xs given path cost qs . Network chooses a price for each link pl based on congestion. Distributed optimisation. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 5/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Background Price based rate control [Kelly97] developed an optimisation model to investigate pricing on broadband networks. Network and User sub-problems User chooses a willingness to pay ws based on a price. Network chooses source rates xs based on ws Low and Lapsley [Low99] solve the same optimisation problem in a different way: User choose a send rate xs given path cost qs . Network chooses a price for each link pl based on congestion. Distributed optimisation. Since this work NUM has been applied to a many network protocols (see [Chiang07]) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 5/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Background Basic NUM max X Us (xs ) subject to Rx ≤ C s David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 6/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Background Basic NUM max X Us (xs ) subject to Rx ≤ C s Maximise the sum of all source utilities, subject to capacity U(xs ) xs David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 6/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Each source, s, solves: xs = (Us0 )−1 (qs ), where qs = X rls pl , and Us its utility. l Note R = rls , the route matrix, rls = 1 if source s uses link l. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 7/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Each source, s, solves: xs = (Us0 )−1 (qs ), where qs = X rls pl , and Us its utility. l Network sub-problem ∂L(x, p) X = rls xs − cl = 0 ∂pl s ∀l. Price pl = 0 if no congestion and pl > 0 if congestion. Note R = rls , the route matrix, rls = 1 if source s uses link l. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 7/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Mo’s and Warland’s α-fair utilities Mo’s and Warland’s [Mo00] α-fair utility family: xs = U 0 (qs ) = qs−α David Hayes david.hayes@ieee.org Rate Control with Packet Corruption (1) 8/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Mo’s and Warland’s α-fair utilities Mo’s and Warland’s [Mo00] α-fair utility family: xs = U 0 (qs ) = qs−α (1) Where: α = 1 yields Kelly’s proportional, fairness [Kelly98], David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 8/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Mo’s and Warland’s α-fair utilities Mo’s and Warland’s [Mo00] α-fair utility family: xs = U 0 (qs ) = qs−α (1) Where: α = 1 yields Kelly’s proportional, fairness [Kelly98], α → ∞ yields max-min fairness, David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 8/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Mo’s and Warland’s α-fair utilities Mo’s and Warland’s [Mo00] α-fair utility family: xs = U 0 (qs ) = qs−α (1) Where: α = 1 yields Kelly’s proportional, fairness [Kelly98], α → ∞ yields max-min fairness, α → 0 sources become less sensitive to price changes, David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 8/ 33 Problem and Objectives Background NUM with packet corruption Examples ICCRG and IETF Price based rate control Network Utility Maximisation Framework α-fair utilities Mo’s and Warland’s α-fair utilities Mo’s and Warland’s [Mo00] α-fair utility family: xs = U 0 (qs ) = qs−α (1) Where: α = 1 yields Kelly’s proportional, fairness [Kelly98], α → ∞ yields max-min fairness, α → 0 sources become less sensitive to price changes, α = 2 approximates TCP fairness [Massoulie02]. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 8/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption A traffic source, s, sending data at rate xs has a fraction s corrupted by the network. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 9/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption A traffic source, s, sending data at rate xs has a fraction s corrupted by the network. The utility of source s is now: U(xs ; s ) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 9/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption A traffic source, s, sending data at rate xs has a fraction s corrupted by the network. The utility of source s is now: U(xs ; s ) If corrupt packets contain no information Source s, transmitting at rate xs only achieves utility for xs (1 − s ): U(xs ; s ) = U(xs (1 − s ); 0). David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 9/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities Some assumptions Fluid flow (general NUM). David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 10/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities Some assumptions Fluid flow (general NUM). Sources see Link price in proportion to their rate (general NUM) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 10/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities Some assumptions Fluid flow (general NUM). Sources see Link price in proportion to their rate (general NUM) pl = 0 when link underutilised, pl > 0 when congestion. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 10/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities Some assumptions Fluid flow (general NUM). Sources see Link price in proportion to their rate (general NUM) pl = 0 when link underutilised, pl > 0 when congestion. Probability of corruption is independent of congestion. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 10/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities Some assumptions Fluid flow (general NUM). Sources see Link price in proportion to their rate (general NUM) pl = 0 when link underutilised, pl > 0 when congestion. Probability of corruption is independent of congestion. Greedy sources. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 10/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption If corrupt packets travel through the network and use capacity David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 11/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption If corrupt packets travel through the network and use capacity Calculation of pl will not change David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 11/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption If corrupt packets travel through the network and use capacity Calculation of pl will not change It follows then: xs = (Us0 )−1 (qs ; s ) 1 qs 0 −1 = (U ) ;0 . 1 − s 1 − s David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 11/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption If the network discards corrupt packets David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 12/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption If the network discards corrupt packets The network calculates pl based on the thinned traffic. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 12/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption If the network discards corrupt packets The network calculates pl based on the thinned traffic. The path cost for s, is now a weighted sum of pl David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 12/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption If the network discards corrupt packets The network calculates pl based on the thinned traffic. The path cost for s, is now a weighted sum of pl Let r̂ls be the fraction of s’s data that traverses link l. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 12/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption If the network discards corrupt packets The network calculates pl based on the thinned traffic. The path cost for s, is now a weighted sum of pl Let r̂ls be the fraction of s’s data that traverses link l. P The sum of link prices for s is now q̂s = l r̂ls pl and David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 12/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption If the network discards corrupt packets The network calculates pl based on the thinned traffic. The path cost for s, is now a weighted sum of pl Let r̂ls be the fraction of s’s data that traverses link l. P The sum of link prices for s is now q̂s = l r̂ls pl and q̂s 1 0 −1 0 −1 xs = (U ) (q̂s ; s ) = (U ) ;0 . 1 − s 1 − s David Hayes david.hayes@ieee.org Rate Control with Packet Corruption (2) 12/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption If the network discards corrupt packets The network calculates pl based on the thinned traffic. The path cost for s, is now a weighted sum of pl Let r̂ls be the fraction of s’s data that traverses link l. P The sum of link prices for s is now q̂s = l r̂ls pl and q̂s 1 0 −1 0 −1 xs = (U ) (q̂s ; s ) = (U ) ;0 . 1 − s 1 − s (2) But how does s calculate q̂s ? David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 12/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption If the network discards corrupt packets The network calculates pl based on the thinned traffic. The path cost for s, is now a weighted sum of pl Let r̂ls be the fraction of s’s data that traverses link l. P The sum of link prices for s is now q̂s = l r̂ls pl and q̂s 1 0 −1 0 −1 xs = (U ) (q̂s ; s ) = (U ) ;0 . 1 − s 1 − s (2) But how does s calculate q̂s ? How does s know R̂ = r̂ls ? David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 12/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities NUM with packet corruption If the network discards corrupt packets The network calculates pl based on the thinned traffic. The path cost for s, is now a weighted sum of pl Let r̂ls be the fraction of s’s data that traverses link l. P The sum of link prices for s is now q̂s = l r̂ls pl and q̂s 1 0 −1 0 −1 xs = (U ) (q̂s ; s ) = (U ) ;0 . 1 − s 1 − s (2) But how does s calculate q̂s ? How does s know R̂ = r̂ls ? If s only knows R will xs be very different? David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 12/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities Using Mo’s and Warland’s α-fair utility (1) with (2): xs = (1 − s )1/α−1 qˆs −1/α David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 13/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities Using Mo’s and Warland’s α-fair utility (1) with (2): xs = (1 − s )1/α−1 qˆs −1/α Interpretation Max-min fairness, α → ∞ Sources with a lower throughput due to s have xs boosted. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 13/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities Using Mo’s and Warland’s α-fair utility (1) with (2): xs = (1 − s )1/α−1 qˆs −1/α Interpretation Max-min fairness, α → ∞ Sources with a lower throughput due to s have xs boosted. Maximum throughput, α → 0 sources with the lowest s receive all the bandwidth. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 13/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities Using Mo’s and Warland’s α-fair utility (1) with (2): xs = (1 − s )1/α−1 qˆs −1/α Interpretation Max-min fairness, α → ∞ Sources with a lower throughput due to s have xs boosted. Maximum throughput, α → 0 sources with the lowest s receive all the bandwidth. Proportional fairness, α = 1 xs is independent of s . David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 13/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities Using Mo’s and Warland’s α-fair utility (1) with (2): xs = (1 − s )1/α−1 qˆs −1/α Interpretation Max-min fairness, α → ∞ Sources with a lower throughput due to s have xs boosted. Maximum throughput, α → 0 sources with the lowest s receive all the bandwidth. Proportional fairness, α = 1 xs is independent of s . If corrupt packets use network capacity q̂s = qs David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 13/ 33 Problem and Objectives Background NUM with packet corruption Examples Case 1: Corrupt packets use network capacity Case 2: Corrupt packets are discarded Using Mo’s and Warland’s α-fair utilities Using Mo’s and Warland’s α-fair utility (1) with (2): xs = (1 − s )1/α−1 qˆs −1/α Interpretation Max-min fairness, α → ∞ Sources with a lower throughput due to s have xs boosted. Maximum throughput, α → 0 sources with the lowest s receive all the bandwidth. Proportional fairness, α = 1 xs is independent of s . If corrupt packets use network capacity q̂s = qs Simple implementation similar to HS-TCP and H-TCP ŵ = (1 − )1/α−1 w David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 13/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Examples Example 1 - Simple 4 link network link 1 link 3 1 2 3 4 1 3 link 2 link 4 2 4 Figure: Simple 4 link, 4 source network David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 14/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Examples Example 1 - Simple 4 link network link 1 link 3 1 2 3 4 1 3 link 2 link 4 2 4 Figure: Simple 4 link, 4 source network Models four different paths, but all are two hops. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 14/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Examples Example 1 - Simple 4 link network link 1 link 3 1 2 3 4 1 3 link 2 link 4 2 4 Figure: Simple 4 link, 4 source network Models four different paths, but all are two hops. Look at: Corruption on link 1 and link 3 with half capacity Corruption on link 3 and link 1 with half capacity David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 14/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Corruption loss on link 1 1.8 1 2 1.6 3 4 Transmission rate 1.4 2 1 1 3 2 2 2 4 source 1 source 2 source 3 source 4 source 1 at link 3 source 2 at link 4 1.2 1 0.8 0.6 0.4 0.2 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 1 0.35 0.4 0.45 Figure: Sources ignore loss due to corruption (α = 2) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 15/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Corruption loss on link 1 1.8 1 2 1.6 3 4 Transmission rate 1.4 2 1 1 3 2 2 2 4 source 1 source 2 source 3 source 4 source 1 at link 3 source 2 at link 4 1.2 1 0.8 0.6 0.4 0.2 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 1 0.35 0.4 0.45 Figure: Sources adapt to corruption (α = 2) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 16/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Corruption loss on link 3 1.8 1 2 1.6 3 4 Transmission rate 1.4 1 2 1 3 2 2 source 1 source 2 source 3 source 4 2 4 1.2 1 0.8 0.6 0.4 0.2 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 3 0.35 0.4 0.45 Figure: Sources ignore loss due to corruption (α = 2) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 17/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Corruption loss on link 3 1.8 1 2 1.6 3 4 Transmission rate 1.4 1 2 1 3 2 2 source 1 source 2 source 3 source 4 2 4 1.2 1 0.8 0.6 0.4 0.2 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 3 0.35 0.4 0.45 Figure: Sources adapt to corruption (α = 2) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 18/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Corruption loss on link 3 3.2 Full corruption capability Sources ignoring corruption 3 Rx rate 2.8 2.6 2.4 2.2 2 1.8 0 0.1 0.2 0.3 0.4 Error rate on link 3 0.5 0.6 Figure: Total received throughput (α = 2) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 19/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion α = 2 fairness Sources experiencing corruption have reduced utility. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 20/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion α = 2 fairness Sources experiencing corruption have reduced utility. Therefore xs is relatively higher than s = 0 sources. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 20/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion α = 2 fairness Sources experiencing corruption have reduced utility. Therefore xs is relatively higher than s = 0 sources. Network is efficient when corruption is on first hop David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 20/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion α = 2 fairness Sources experiencing corruption have reduced utility. Therefore xs is relatively higher than s = 0 sources. Network is efficient when corruption is on first hop Network less efficient when corruption is on last hop David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 20/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion α = 2 fairness Sources experiencing corruption have reduced utility. Therefore xs is relatively higher than s = 0 sources. Network is efficient when corruption is on first hop Network less efficient when corruption is on last hop Is α = 2 fairness the right balance between end-user utility and network resource efficiency? David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 20/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion α = 2 fairness Sources experiencing corruption have reduced utility. Therefore xs is relatively higher than s = 0 sources. Network is efficient when corruption is on first hop Network less efficient when corruption is on last hop Is α = 2 fairness the right balance between end-user utility and network resource efficiency? Setting α < 2 shifts the balance toward network efficiency David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 20/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Corruption loss on link 3 1.8 1 2 1.6 3 4 Transmission rate 1.4 1 2 1 3 2 2 source 1 source 2 source 3 source 4 2 4 1.2 1 0.8 0.6 0.4 0.2 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 3 0.35 0.4 0.45 Figure: Sources adapt to corruption (α = 0.8) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 21/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Comparing the characteristics of α = 2 with α = 0.8 1.8 1.8 1 2 1.6 3 4 2 1 3 2 2 source 1 source 2 source 3 source 4 1 2 1.6 2 4 3 4 1.4 Transmission rate Transmission rate 1.4 1 1.2 1 0.8 0.6 0.4 1 2 1 3 2 2 source 1 source 2 source 3 source 4 2 4 1.2 1 0.8 0.6 0.4 0.2 0.2 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 3 0.35 0.4 0.45 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 3 0.35 0.4 0.45 Figure: Sources adapt to corruption (α = 2) and (α = 0.8) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 22/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Comparing the characteristics of α = 2 with α = 0.8 1.8 1.8 1 2 1.6 3 4 2 1 3 2 2 source 1 source 2 source 3 source 4 1 2 1.6 2 4 3 4 1.4 Transmission rate Transmission rate 1.4 1 1.2 1 0.8 0.6 0.4 1 2 1 3 2 2 source 1 source 2 source 3 source 4 2 4 1.2 1 0.8 0.6 0.4 0.2 0.2 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 3 0.35 0.4 0.45 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 3 0.35 0.4 0.45 Figure: Sources adapt to corruption (α = 2) and (α = 0.8) High α → high throughput fairness Low α → low throughput fairness David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 22/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Corruption loss on link 3 3 Full corruption capability Sources ignoring corruption 2.9 Rx rate 2.8 2.7 2.6 2.5 2.4 2.3 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 3 0.35 0.4 0.45 Figure: Total received throughput (α = 0.8) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 23/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Corruption loss on link 1 1.8 1 2 1.6 3 4 Transmission rate 1.4 2 1 1 3 2 2 2 4 source 1 source 2 source 3 source 4 source 1 at link 3 source 2 at link 4 1.2 1 0.8 0.6 0.4 0.2 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 1 0.35 0.4 0.45 Figure: Sources adapt to corruption (α = 0.8) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 24/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Comparing the characteristics of α = 2 with α = 0.8 1.8 1.8 1 2 1.6 3 4 1 1 3 2 2 2 4 source 1 source 2 source 3 source 4 source 1 at link 3 source 2 at link 4 1.2 1 0.8 0.6 1 2 1.6 3 4 1.4 Transmission rate Transmission rate 1.4 2 0.4 2 1 2 2 1 3 2 4 source 1 source 2 source 3 source 4 source 1 at link 3 source 2 at link 4 1.2 1 0.8 0.6 0.4 0.2 0.2 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 1 0.35 0.4 0.45 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 1 0.35 0.4 0.45 Figure: Sources adapt to corruption (α = 2) and (α = 0.8) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 25/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion Interestingly the case with corruption on the first link behaves similarly for both α = 2 and α = 0.8 David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 26/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion Interestingly the case with corruption on the first link behaves similarly for both α = 2 and α = 0.8 Increasing the send rate of sources 1 and 2 also increases the network efficiency in both cases. Corrupt packets are dropped at link 1. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 26/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion Interestingly the case with corruption on the first link behaves similarly for both α = 2 and α = 0.8 Increasing the send rate of sources 1 and 2 also increases the network efficiency in both cases. Corrupt packets are dropped at link 1. Also, this is a special case where all flows have two hops. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 26/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion Interestingly the case with corruption on the first link behaves similarly for both α = 2 and α = 0.8 Increasing the send rate of sources 1 and 2 also increases the network efficiency in both cases. Corrupt packets are dropped at link 1. Also, this is a special case where all flows have two hops. A source’s utility is calculated using sum of link prices along its path. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 26/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion Interestingly the case with corruption on the first link behaves similarly for both α = 2 and α = 0.8 Increasing the send rate of sources 1 and 2 also increases the network efficiency in both cases. Corrupt packets are dropped at link 1. Also, this is a special case where all flows have two hops. A source’s utility is calculated using sum of link prices along its path. This the dynamic changes when sources travel different numbers of hops. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 26/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Example 2 - Simple 4 link 8 source network 5 5 7 link 1 7 link 3 1 2 1 3 link 2 3 4 6 link 4 6 8 2 4 8 Figure: Simple 4 link, 8 source network David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 27/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Example 2 - Simple 4 link 8 source network 5 5 7 link 1 7 link 3 1 2 1 3 link 2 3 4 6 link 4 6 8 2 4 8 Figure: Simple 4 link, 8 source network Models eight different paths of 1 or 2 hop lengths. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 27/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Example 2 - Simple 4 link 8 source network 5 5 7 link 1 7 link 3 1 2 1 3 link 2 3 4 6 link 4 6 8 2 4 8 Figure: Simple 4 link, 8 source network Models eight different paths of 1 or 2 hop lengths. Look at: Corruption on link 1 and link 3 with half capacity Corruption on link 3 and link 1 with half capacity David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 27/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Corruption loss on link 1 (subset - link 1) 1.8 5 7 1.5 7 1 3 3 4 1.4 2 4 6 Transmission rate 5 3 1 2 1.6 3 6 8 3 8 1 2 5 1 at 3 2 at 4 5 at 1 1.2 1 0.8 0.6 0.4 0.2 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 1 0.35 0.4 0.45 Figure: Sources adapts to corruption (α = 2) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 28/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion Generally even distribution of capacity on all links. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 29/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion Generally even distribution of capacity on all links. Single hop flows have a slight advantage. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 29/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion Generally even distribution of capacity on all links. Single hop flows have a slight advantage. on link 1 deceases the Us of all flows on the link. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 29/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion Generally even distribution of capacity on all links. Single hop flows have a slight advantage. on link 1 deceases the Us of all flows on the link. However, only the two hop flows increase their send rates. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 29/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion Generally even distribution of capacity on all links. Single hop flows have a slight advantage. on link 1 deceases the Us of all flows on the link. However, only the two hop flows increase their send rates. Flows 1 and 2 were restricted by the second hop. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 29/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion Generally even distribution of capacity on all links. Single hop flows have a slight advantage. on link 1 deceases the Us of all flows on the link. However, only the two hop flows increase their send rates. Flows 1 and 2 were restricted by the second hop. When link 1 becomes the bottleneck for all three sources link 1 sources share capacity evenly. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 29/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Corruption loss on link 3 5 5 7 1.5 1 2 7 3 1 3 3 4 1.8 2 4 6 3 6 8 3 1 2 3 4 5 6 7 8 8 1.6 Transmission rate 1.4 1.2 1 0.8 0.6 0.4 0.2 0 0.05 0.1 0.15 0.2 0.25 0.3 Error rate on link 3 0.35 0.4 0.45 Figure: Utility adapts to corruption (α = 2) David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 30/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion 2 hop sources with s > 0 lift their relative rates. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 31/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion 2 hop sources with s > 0 lift their relative rates. 1 hop source, 7, decreases its rate, so as to fairly share link 3. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 31/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion 2 hop sources with s > 0 lift their relative rates. 1 hop source, 7, decreases its rate, so as to fairly share link 3. Single hop flows have a slight advantage. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 31/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion 2 hop sources with s > 0 lift their relative rates. 1 hop source, 7, decreases its rate, so as to fairly share link 3. Single hop flows have a slight advantage. Generally – flows experiencing corruption increase their send rate David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 31/ 33 Problem and Objectives Background NUM with packet corruption Examples Example 1 - Simple 4 link 4 source network Example 2 - Simple 4 link 8 source network Discussion 2 hop sources with s > 0 lift their relative rates. 1 hop source, 7, decreases its rate, so as to fairly share link 3. Single hop flows have a slight advantage. Generally – flows experiencing corruption increase their send rate more fairly shares the “goodput” with their competing flows. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 31/ 33 Problem and Objectives Background NUM with packet corruption Examples Conclusions and further work Traffic sources should adapt to network corruption fairly. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 32/ 33 Problem and Objectives Background NUM with packet corruption Examples Conclusions and further work Traffic sources should adapt to network corruption fairly. TCP and DCCP? David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 32/ 33 Problem and Objectives Background NUM with packet corruption Examples Conclusions and further work Traffic sources should adapt to network corruption fairly. TCP and DCCP? This work gives a simple way of determining what the fair rate should be. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 32/ 33 Problem and Objectives Background NUM with packet corruption Examples Conclusions and further work Traffic sources should adapt to network corruption fairly. TCP and DCCP? This work gives a simple way of determining what the fair rate should be. Can be implemented simply by modifying TCP windows David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 32/ 33 Problem and Objectives Background NUM with packet corruption Examples Conclusions and further work Traffic sources should adapt to network corruption fairly. TCP and DCCP? This work gives a simple way of determining what the fair rate should be. Can be implemented simply by modifying TCP windows At the moment there isn’t agreement in the ICCRG about the user utility/network efficiency balance. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 32/ 33 Problem and Objectives Background NUM with packet corruption Examples Conclusions and further work Traffic sources should adapt to network corruption fairly. TCP and DCCP? This work gives a simple way of determining what the fair rate should be. Can be implemented simply by modifying TCP windows At the moment there isn’t agreement in the ICCRG about the user utility/network efficiency balance. Depends on: What service users are paying for. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 32/ 33 Problem and Objectives Background NUM with packet corruption Examples Conclusions and further work Traffic sources should adapt to network corruption fairly. TCP and DCCP? This work gives a simple way of determining what the fair rate should be. Can be implemented simply by modifying TCP windows At the moment there isn’t agreement in the ICCRG about the user utility/network efficiency balance. Depends on: What service users are paying for. What their use costs network operators. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 32/ 33 Problem and Objectives Background NUM with packet corruption Examples Conclusions and further work Traffic sources should adapt to network corruption fairly. TCP and DCCP? This work gives a simple way of determining what the fair rate should be. Can be implemented simply by modifying TCP windows At the moment there isn’t agreement in the ICCRG about the user utility/network efficiency balance. Depends on: What service users are paying for. What their use costs network operators. IETF politics. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 32/ 33 Problem and Objectives Background NUM with packet corruption Examples If α > 1 fairness may need cap for extreme corruption loss. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 33/ 33 Problem and Objectives Background NUM with packet corruption Examples If α > 1 fairness may need cap for extreme corruption loss. Packet corruption information? David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 33/ 33 Problem and Objectives Background NUM with packet corruption Examples If α > 1 fairness may need cap for extreme corruption loss. Packet corruption information? Lower layer indications David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 33/ 33 Problem and Objectives Background NUM with packet corruption Examples If α > 1 fairness may need cap for extreme corruption loss. Packet corruption information? Lower layer indications Indication packets David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 33/ 33 Problem and Objectives Background NUM with packet corruption Examples If α > 1 fairness may need cap for extreme corruption loss. Packet corruption information? Lower layer indications Indication packets Congestion indication — loss only from corruption David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 33/ 33 Problem and Objectives Background NUM with packet corruption Examples If α > 1 fairness may need cap for extreme corruption loss. Packet corruption information? Lower layer indications Indication packets Congestion indication — loss only from corruption Will qs be a good enough approximation of qˆs David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 33/ 33 Problem and Objectives Background NUM with packet corruption Examples If α > 1 fairness may need cap for extreme corruption loss. Packet corruption information? Lower layer indications Indication packets Congestion indication — loss only from corruption Will qs be a good enough approximation of qˆs Models for burstiness, TCP-newRENO specifics, real-time traffic? David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 33/ 33 Bibliography M. Chiang, S. H. Low, A. R. Calderbank, and J. C. Doyle. Layering as optimization decomposition: A mathematical theory of network architectures. Proc. IEEE, 95(1):255–312, January 2007. Frank Kelly. Charging and rate control for elastic traffic. European Transactions on Telecommunications, 8:33–37, 1997. Frank Kelly, Aman Maulloo, and David Tan. Rate control in communication networks: Shadow prices, proportional fairness and stability. J. Op. Res. Soc., 49:237–378, 1998. S. H. Low and D. E. Lapsley. Optimization flow control I: Basic algorithm and convergence. IEEE/ACM Trans. Networking, 7(6):861–875, December 1999. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 33/ 33 Bibliography L. Massoulie and J. Roberts. Bandwidth sharing: objectives and algorithms. IEEE/ACM Trans. Networking, 10(3):320–328, June 2002. J. Mo and J. Walrand. Fair end-to-end window-based congestion control. IEEE/ACM Trans. Networking, 8(5):556–567, October 2000. David Hayes david.hayes@ieee.org Rate Control with Packet Corruption 33/ 33