NUS.SOC.CS5248 Ooi Wei Tsang Previously, on CS5248..

NUS.SOC.CS5248 Ooi Wei Tsang IP Multicast

NUS.SOC.CS5248 Ooi Wei Tsang New Model: IP Multicast sender receivers

NUS.SOC.CS5248 Ooi Wei Tsang Question 1 Router G ? should I forward this packet to my subnet?

NUS.SOC.CS5248 Ooi Wei Tsang IGMP v2.0 JOIN message A : “I want to join group G.” QUERY message R : “Which group have you joined ?”

NUS.SOC.CS5248 Ooi Wei Tsang IGMP v2.0 LEAVE message “I want to leave group G” Group-Specific Query “Anybody else belongs to group G ?”

NUS.SOC.CS5248 Ooi Wei Tsang Rate Adaptations

NUS.SOC.CS5248 Ooi Wei Tsang Layered Video Layer 1 Layer 2Layer 3

NUS.SOC.CS5248 Ooi Wei Tsang TCP-Equation Window size behavior in TCP/IP with constant loss probability T. Ott, J. Kemperman, and M. Mathis June 1997, HPCS 1997

NUS.SOC.CS5248 Ooi Wei Tsang Heterogeneity

NUS.SOC.CS5248 Ooi Wei Tsang Internet Heterogeneity 2 Mbps 40kbps 56kbps

NUS.SOC.CS5248 Ooi Wei Tsang Two Simple Solutions one stream to fit them all send multiple streams

NUS.SOC.CS5248 Ooi Wei Tsang Network Encoder Sender Middlebox Receiver Decoder

NUS.SOC.CS5248 Ooi Wei Tsang Receiver-Driven Layered Multicast Steve McCanne, Van Jacobson, Martin Vetterli SIGCOMM 1996

NUS.SOC.CS5248 Ooi Wei Tsang Layered Multicast 1 Layer : 1 Multicast Group Receiver subscribes to as many layers as desired

NUS.SOC.CS5248 Ooi Wei Tsang RLM Example SR1 R2 R3R4

NUS.SOC.CS5248 Ooi Wei Tsang Questions How many layers are enough?

NUS.SOC.CS5248 Ooi Wei Tsang Solution: Join Experiment highest layer = 1 join layer 1 while no packet loss highest layer ++ join next layer leave highest layer highest layer --

NUS.SOC.CS5248 Ooi Wei Tsang Details T join Time between join experiments T detect Time taken to detect packet loss

NUS.SOC.CS5248 Ooi Wei Tsang Effects of T join Need to converge to the right level quickly T join should be small Repeated failed experiments congest networks T join should be large

NUS.SOC.CS5248 Ooi Wei Tsang Adapting T join (i) One T join per layer if join experiment for layer k fails T join (k) = T join (k)*2

NUS.SOC.CS5248 Ooi Wei Tsang Example

NUS.SOC.CS5248 Ooi Wei Tsang Adapting T detect Set T detect to large initial value Estimate T detect with mean and deviation Measure time between join and packet loss occur

NUS.SOC.CS5248 Ooi Wei Tsang Two Problems Interference Scalability

NUS.SOC.CS5248 Ooi Wei Tsang Problem 1: Interference

NUS.SOC.CS5248 Ooi Wei Tsang Problem 1: Interference I see, layer 2 is bad for me..

NUS.SOC.CS5248 Ooi Wei Tsang Problem 2: Scalability Lots of receivers Lots of experiments Lots of congestions

NUS.SOC.CS5248 Ooi Wei Tsang Solution: Shared Learning I am joining layer 2, do not disturb!

NUS.SOC.CS5248 Ooi Wei Tsang Solution: Shared Learning

NUS.SOC.CS5248 Ooi Wei Tsang Solution: Shared Learning I am joining layer 3, do not disturb!

NUS.SOC.CS5248 Ooi Wei Tsang Solution: Shared Learning I see, layer 3 is bad for me..

NUS.SOC.CS5248 Ooi Wei Tsang Solution: Shared Learning I see, layer 3 is bad for me..

NUS.SOC.CS5248 Ooi Wei Tsang Problem Reducing interference decrease convergence time Solution:

NUS.SOC.CS5248 Ooi Wei Tsang Evaluation Session Size Convergence Time (s)

NUS.SOC.CS5248 Ooi Wei Tsang Evaluation Loss Rate (%) Session Size 1 100

NUS.SOC.CS5248 Ooi Wei Tsang Problems with RLM Not TCP Friendly Share learning is “clumsy” Failed join experiments is bad for the network

NUS.SOC.CS5248 Ooi Wei Tsang Thin Streams: An Architecture for Multicasting Layered Video L. Wu, R. Sharma., B. Smith NOSSDAV 97

NUS.SOC.CS5248 Ooi Wei Tsang How bad is failed experiment? R: sending rate of a layer T j : IGMP join latency T l : IGMP leave latency T detect : time before drop Buffer space at the router =

NUS.SOC.CS5248 Ooi Wei Tsang How big is IGMP Leave Latency?

NUS.SOC.CS5248 Ooi Wei Tsang What To Do? Buffer space at the router = R(T j + T l + T detect ) Proposal of ThinStream Reduce R Reduce T detect

NUS.SOC.CS5248 Ooi Wei Tsang Reducing T detect Detect congestion before it happens

NUS.SOC.CS5248 Ooi Wei Tsang Calculating A & E R: bandwidth of one layer I: measurement interval N: number of bytes received in I G: number of layers joined A = E =

NUS.SOC.CS5248 Ooi Wei Tsang ThinStreams Algorithm if (E - A > leave_threshold) leave else if (at least T join since last join) and (E - A < join_threshold) join

NUS.SOC.CS5248 Ooi Wei Tsang Fairness Improve fairness leave_threshold as a function of G

NUS.SOC.CS5248 Ooi Wei Tsang Problems with ThinStream Small R Still not TCP friendly

NUS.SOC.CS5248 Ooi Wei Tsang TCP-Like Congestion Control for Layered Multicast Data Transfer L. Vicisano, J. Crowcroft, L. Rizzo INFOCOM 98

NUS.SOC.CS5248 Ooi Wei Tsang The RLC Protocol 1. Try to emulate TCP 2. Synchronize join 3. “Probe” before join

NUS.SOC.CS5248 Ooi Wei Tsang Recall TCP’s Behavior AIMD

NUS.SOC.CS5248 Ooi Wei Tsang Recall Layered Multicast

NUS.SOC.CS5248 Ooi Wei Tsang 1. Emulating TCP Idea: By finding appropriate B i and T join (i)

NUS.SOC.CS5248 Ooi Wei Tsang Idea

NUS.SOC.CS5248 Ooi Wei Tsang 2. Synchronizing Join Idea: send special packets in data stream to signal join Distance between synchronization point for layer i is T join (i)

NUS.SOC.CS5248 Ooi Wei Tsang Sync Point

NUS.SOC.CS5248 Ooi Wei Tsang 3. Probe before Join Emulate joining the next layer If this causes packet drop, don’t join

NUS.SOC.CS5248 Ooi Wei Tsang Sender-Based Probing base layer i

NUS.SOC.CS5248 Ooi Wei Tsang Sender-Based Probing base layer i

NUS.SOC.CS5248 Ooi Wei Tsang Problems with RLC Burst too short No fine grain increment

NUS.SOC.CS5248 Ooi Wei Tsang Summary Receiver Driven Layered Multicast Lots of Problems No Perfect Solutions