Realtime Multimedia Streaming over Internet Pengjun Pei Dazhen

Realtime Multimedia Streaming over Internet Pengjun Pei Dazhen Pan CSE 620 Fall, 2001

Overview Wide Application n Video-conference Internet telephony Streaming audio/video players Challenges: Internet is best-effort network n n n Packet loss Bandwidth variation Packet delay variation

System Architecture

Content Video Compression Congestion Control Error Control

Video Compression Various requirement: n n n Bandwidth Delay Loss VCR like function Decoding complexity Intra-frame redundancy & inter-frame redundancy Non-scalable coding vs Scalable coding

Inter-frame redundancy MPEG-2: I frame: intra-picture P frame: predicted picture B frame: bi-directional predicted picture MPEG frame dependencies in an MPEG bit stream

Scalable Coding FGS: fine granularity scalability(proposed to MPEG-4): Bitplanes of enhancement DCT coeffients

Content Video Compression Congestion Control Error Control

Congestion Control Requirements for multimedia streaming n n Relatively constant rate Low latency for packet delivery Small latency variance Timely delivery is more important than complete reliability Rate control Rate shaping

TCP/UDP? TCP n n Retransmission mechanism intolerable delays Multiplicative decrease in case of congestion sharp variation in visual effect UDP n n Unfair to responsive TCP flows Congestion collapse

Categories of Rate Control Source-based rate control Source adjusts sending rate Feedback employed Can be applied to both unicast and multicast Receiver-based rate control Receiver joins layer/channel Used in multicasting scalable video Hybrid rate control

TCP Friendly Flows A flow is TCP-friendly if its arrival rate does not exceed the bandwidth of a conformant TCP connection in the same circumstances.

TCP throughput model λ: MTU: RTT: p: throughput of a TCP connection maximum transit unit Round Trip Time packet loss ratio

RAP(Rate Adaptation Protocol) Proposed by R. Rejaie 1998 End-to-end architecture

RAP Decision function n n If no congestion is detected, periodically increase the transmission rate If congestion is detected, immediately decrease the transmission rate Increase/Decrease algorithm: AIMD Decision frequency n Smoothed version of one RTT: most recent value of SRTT

RAP Decision function Mechanisms to detect loss: n Timeout SRTTi = 7/8 * SRTTi + 1/8 * Sample. RTT Timeout=μ*SRTT+δ*Var. RTT Use transmission history ’coz it isn’t ack-clocked Before sending a new packet, source traverses through the transmission history and detects all timeout losses: WHILE (Depart. Timei+Timeout>=Curr. Time) IF(Flagi!=Acked) THEN Seqi is lost Detect a burst of loss at once

RAP Decision function (Continued) n Gaps in sequence number(ACK-based) ACK Packet: Acurr: packet being acknowledged N: the last packet before Acurr that was still missing Alast: the last packet before N that was received n Timeout mechanism as a backup for critical scenarios such as when a burst of packets is lost

AIMD in RAP No-packet loss: n n Si = Si + α (step height) Si = Packet. Size/IPGi+1 = IPGi*C/( IPGi + C ) α = Si+1 – Si = Packet. Size/C Upon packet loss: n n n Si+1 = β*Si IPGi+1 = IPGi/β β = 0. 5 IPG: inter-packet-gap

RAP Decision Frequency Adjust IPG once every round-trip time using most recent value of SRTT Right value of C: n n n C must be adjusted so that in a steady state, the number of packets transmitted per step is increased by 1. If IPG is updated once every T seconds and we choose C = T/k, the # of packets sent during each step is increased by k every step. RAP use k=1 to emulate the TCP window adjust

RAP Fine-grain rate adaptation Motivation: Make RAP more stable and responsive to transient congestion while still performing the AIMD algorithm at a coarser granularity Fine-grain feedback: Feedbacki=FRTTi/XRTTi FRTTi, XRTTi: short/long term exponential moving average of RTT samples at the ith adjusting point RTTi+1 = (1 – K)RTTi+K*Sample. RTT (KXRTT=0. 01 KFRTT=0. 9) Fine-grain adjustment IPGi’ = IPGi * Feedbacki

Simulation Result RAP

Simulation Result(FG-RAP)

Binomial Algo Proposed by D. Bansal 2000 I: Increase in window as a result of receipt of one window of ACK in a RTT D: Decrease in window on detection of a loss by the sender Wt: window size at time t

Properties of Binomial Algo Any l < 1 has a decrease that is in general less than a multiplicative decrease TCP Friendly if and only if k + l = 1 and l <= 1 for suitable α and β. Converge to fairness as long as k > =0, l >= 0, k + l > 0

Ratio of throughput AIMD/Binomial x: value of k y: TCP throughput/Binomial throuput

SQRT(k = l = 0. 5)

SQRT vs AIMD SQRT has less oscillatory bandwidth probing

Categories of Rate Control Source-based rate control Source adjusts sending rate Feedback employed Can be applied to both unicast and multicast Receiver-based rate control Receiver joins layer/channel Used in multicasting scalable video Hybrid rate control

Source-based Rate Control for Multicast Unicast video distribution using multiple point -point connection Multicast video distribution using point-tomultipoint transmission

Single-Channel Multicast IVS(INRIA Video-conference System): n n n Single-channel multicast Probe-base, use AIMD Each receiver determine the network status Source solicits network status info through probabilistic polling to avoid feedback implosion Compare the fraction of congested receiver with threshold

Multiple-channel multicast Differentiated service to receivers because each receiver can individually negotiate service parameters with the recourse Bandwidth inefficiency

Categories of Rate Control Source-based rate control Source adjusts sending rate Feedback employed Can be applied to both unicast and multicast Receiver-based rate control Receiver joins layer/channel Used in multicasting scalable video Hybrid rate control

Receiver-based Rate Control Typically applied to layered multicast video n n Source-based works reasonably well for unicast Receiver-based targeted at solving heterogeneity problem in the multicast case Probe-based: n n No congestion, receiver probes for available bandwidth by joining layer/channel When congested, receiver drops a layer

Receiver-based Rate Control Model-based: n n n Based on throughput model of TCP γi: transmission rate of Layer I, L current highest layer Starts with subscribing base layer(Layer 0), set L=0. Obtain MTU, RTT, p for a given period, calculate throughput λ. If λ < γ 0 drop base layer and stop receiving video Else determine L’ , the largest integer such that

Categories of Rate Control Source-based rate control Source adjusts sending rate Feedback employed Can be applied to both unicast and multicast Receiver-based rate control Receiver joins layer/channel Used in multicasting scalable video Hybrid rate control

Hybrid Rate Control Targeted at multicast video Applicable to both layered video and nonlayered video Multiple channels, sender dynamically adjusts the rate for each channel DSG(Destination Set Grouping) n n n Multiple streams: same video info with different rate and quality, each sent to an IP multicast group Receiver chooses a multicast group to join Source uses feedback to adjust rate for each stream

Rate Shaping Adapt the rate of compressed video bitstreams to the target rate constraint

Types of rate filter Codec Filter: perform transcoding between different schemes Frame-dropping filter: distinguish frame types and drop frames according to importance Layer-dropping filter: distinguish layers and drop frames according to importance Frequency filter: discard DCT coefficient of high frequencies

Conclusion Binomial rate control causes less oscillation to multimedia stream Current research separates rate control and rate shaping

Main Reference Deepak Bansal and Hari Balakrishnan , Binomial Congestion Control Proc. IEEE INFOCOM Conf. , Anchorage, AK, April 2001. S. Floyd, M. Handley, J. Padhye, and J. Widmer, Equation-Based Congestion Control for Unicast Applications, Proc. ACM SIGCOMM’ 00, pages 43 -54, Stockholm, Sweden, September 2000 International Organization for Standardization. Overview of the MPEG-4 Standard, December 1999 Sally Floyd, Kevin Fall, Promoting the Use of End-to-End Congestion Control in the Internet IEEE/ACM Transaction on Networking, May 1999 Dapeng Wu, Yiwei Thomas Hou, etc, Streaming Video over the Internet: Approaches and Directions IEEE Transaction on Circuits and System for Video Technology, Vol 11, No 1, February 2001 Dapeng Wu, Yiwei Thomas Hou, etc, Transorting real-time video over the Internet: challenges and approaches, Proceedings of the IEEE, vl. 88, no. 12, Dec. 2000 R. Padhye, J. Kurose, D. Towsley, and R. Koodi. A Model-based TCP-Friendly Rate Control Protocol. In Proc. IEEE NOSSDAV’ 99, Basking Ridge, New Jersey, June 1999 R. Rejaie, M. Handely, and D. Estrin. RAP: An End-to-end Rate-based Congestion Control Mechanism for Realtime Streams in the Internet. In Proc. IEEE Infocom’ 99, New York, NY, March 1999