Optimal Stream Replication for Video Simulcasting

Jiangchuan Liu, Member, IEEE, Bo Li, Senior Member, IEEE, and Ya-Qin Zhang, Fellow, IEEE

IEEE TRANSACTIONS ON MULTIMEDIA, FEBRUARY 2006

Outline

• Introduction

• What is video simulcasting?

• How to improve user satisfaction– Optimization for Fixed Number of Streams

(OptFN)– Joint Optimization for Stream Number and

Bandwidths (OptNB)

• Experimental results

Introduction

• With broadband networks– Real-time video distribution is a important

application• Ex: multicasting

• Consideration:– How to handle user heterogeneity?

• An equal allocation often leads to a waste of bandwidth

Video simulcasting

• Enables a sender to generate replicated streams of different rates

• Problems:– Stream redundancy– Bandwidth mismatches (affects the user satisfaction)

• Goal:– Strike a balance between bandwidth and user

satisfaction• Use a optimal and efficient solution

– to choose an appropriate number of streams– to allocate bandwidth to reduce bandwidth mismatches for all

the receivers

System model

• In overlay multicast networks

• A server– Has higher output bandwidth– uses simulcasting technique to distribute a

set of video programs

• The receiver – has heterogeneous bandwidth

System definitions(1/2)

• Session (s): a video program (including the replications) and the receivers

• Session bandwidth (Ns): bandwidth allocated to the session

• The status of the system (C, P, Ms,t)

• Bandwidth mismatch

– Use relative measure (RM)• Assume a receive bandwidth t, and the bandwidth of its stream is r

C: the maximum outbound bandwidth of the serverP: the total number of sessions Ms,t: the ratio of the receives having bandwidth t in session s

System definitions(2/2)

• : bandwidth allocation for session s

• rs,i : the rate of stream i

=(rs,1, rs,2, …, rs,ls), rs,1 < rs,2 <, …, < rs,ls

ls: the total number of the replicated streams

• Best-matching bandwidth

• How to determine on the server’s side?– intra-session allocation

Intra-session allocation

• Input: – session bandwidth Ns

– receivers’ bandwidth distribution Ms,t

• Output:– Minimum expected relative mismatch (ERM)

TS: the maximum receiver bandwidth in session s

Optimization for Fixed Number of Streams (OptFN)(1/2)

• Assume– The number of streams is fixed to a given K – A total number k(≤ K) streams are generated

• With a total bandwidth n• Bandwidth of stream k is m

• Ex: if k = 1, and 0 < m = n ≤ Ns;

s

k

i iskss

T

t sts

nrmrkl

RttRMMkmn1 ,

,,

)],(,[),,( min1 ,,

sT

mt ts

m

t ts mtRMMtRMMnm ],[]0,[)1,,( ,

1

0 ,

Optimization for Fixed Number of Streams (OptFN)(2/2)

• When stream k is added– Depend on the bandwidth of stream k and k-1

• If bandwidth of stream k = m, and that of stream k-1 = j;

– Minimum ERM is to checking all possible j(=1,2,…m-1)

• The solution to problem OptFN is

sT

mt ts mtRMjtRMMjmDIFF )],(),([),( ,

),()1,,(),,( min1

jmDIFFkjmnkmnmj

),,(min1,1

Kmnss TmNn

Joint Optimization for Stream Number and Bandwidths (OptNB)(1/2)

• Both the number of streams (ls) and their bandwidth (rs,i) are to be optimized

• Given an upper bound of ls

• Minimum ERM is to try ls from 1 to – With session bandwidth n

– Bandwidth of stream ls is m

2/12/81max ss Nl

s

sl

k kslss

T

t sts

nrmr

RttRMMmn1 ,

,

)],(,[),( min1 ,,

maxsl

Joint Optimization for Stream Number and Bandwidths (OptNB)(2/2)

• For some constraint, m<n≤Ns, 1<m ≤min{n,Ts}

• The solution to problem OptNB

),(),(),( min1

jmDIFFjmnmnmj

),(min1,1

mnss TmNn

Evaluation results

• Simulation subjects include:– A fixed number of streams (ExpFN)

• S. McCanne, V. Jacobson, and M. Vetterli, “Receiver-driven layered multicast,” in Proc. ACM SIGCOMM’ 96, Aug. 1996

• J. Liu, B. Li, and Y.-Q. Zhang, “A hybrid adaptation protocol for TCPfriendly layered multicast and its optimal rate allocation, ”in Proc. IEEE INFOCOM’02, Jun. 2002

– Optimization for Fixed Number of Streams (OptFN)– Joint Optimization for Stream Number and

Bandwidths (OptNB)

• Assume both the minimum and the maximum receiver bandwidths are known– Assume the session has 500 receivers– Their bandwidth are distributed in w clusters

• Follow a Gaussian distribution

Effect of session bandwidth(1/2)

• In w = 3

Effect of session bandwidth(2/2)

• In w = 6

Impact of the number of streams

• If K is small, the receivers’ choice is limited and the adaptation is not flexible

Perceived video quality (1/2)

• Use the standard MPEG-4 video encoder with TM-5 rate control• The test sequence is “Foreman (CIF)”• In w = 3

Perceived video quality (2/2)

• In w = 6