1

Network-supported Rate Control Mechanism for Multicast Streaming Media

Kiyohide NAKAUCHI, Hiroyuki MORIKAWA, and Tomonori AOYAMA,School of Engineering, The University of Tokyo

2

Outline Tow layered multicast rate control

scheme Receiver-driven scheme (RLM) Network-supported scheme

Network-supported layered multicast scheme(NLM)

Comparison

3

Introduction In general, rate control for layered

multicast is classified into two classes Receiver-driven scheme

Each receiver makes decisions to add or drop an enhancement layer

Network-supported scheme Some of internal nodes such as routers or

switches adopt priority dropping mechanism which drops lower priority packets belonging to a higher layer, and decide the number of received layers for each receiver

4

Introduction(Cont.) Receiver-driven scheme

Join-experiment Upgrade the number of layer if no

congestion occurs Degrade the number of layer if

congestion occurs Congestion detection

Loss rate threshold

5

Introduction (Cont.) Drawbacks of receiver-driven

scheme No base layer protection Receivers can’t respond to the

congestion quickly because of both the delay for estimation network conditions and detecting congestion(estimation delay) and the delay for leaving a multicast group

6

NLM Architecture NLM system consists of a sender,

receivers, and routers with rate control mechanisms.

Sender A sender hierarchically encodes sending data

of a session to several CBR streams, and sends all streams on a single multicast group

As routers and receivers cannot identify a layer from its multicast address, the number indicating a layer is filled in somewhere in a packet header

7

NLM Architecture Receiver

A receiver joins a single multicast session on which all the layers of the session are transmitted

Layered-Multicast –capable Router (LMR) LMRS are strategically located at boundaries

between the areas with rich bandwidth and those with poor bandwidth or at the points where congestion frequently occurs

8

Overview of Network-supported Rate Control

Session State Table

oif: output interfaceLcur(Ij, Si): the number of layers allowed to be transmitted for session Si at oif Ij

Lmax(Si): the maximum number of layers allowed to be transmitted for session Si

9

Overview of Network-supported Rate Control Rate control for NLM consists of the fo

llowing three key functions Filtering:

A LMR decides which layer to filter based on the current bandwidth allocation or on the request from downstream LMRs

Signaling: A LMR requests upstream ones to filter a nee

dless layer or to transmit a layer without filtering

10

Overview of Network-supported Rate Control

Session control: LMRs keep consistency of session state table

s with on another

11

Overview of Filtering mechanism

DROP

ADD

12

Overview of signaling protocol

13

Filtering algorithm A LMR conducts rate adaptation by filt

ering when it detects congestion occur or been solved at an oif

Whenever a LMR queues a packet at Ij , it calculates the current qlenj using both the current queue length( ) and previous qlenj as folllows

14

In order to wait the effect of the DROP (ADD) to be reflected to qlenj

and Lcur(Ij , Si ) > 1

15

Signaling Protocol

Oif(j) of this LMR received command DRPO_REQ(Si, L)

Oif(j) of this LMR received command ADD_REQ(Si, L)

16

Session control The purpose of session control is to ke

ep consistency of Lmax and Lcur among session state tables of LMRs and to keep correct PrevID

Consistency of session state tables enables a LMR to select the proper target layer of DROP or ADD

17

Session control(Cont.) A sender of Si periodically multicasts SESS(Si , L) at

intervals of ss_intvl when the sender is transmitting L (L = Lcur(Ij, Si)) layers

18

Extend mechanism In NLM, a LMR changes add_intvl j adaptively to th

e network conditions in order to minimize the frequency of rate oscillation

A LMR changes add_intvl j according to the following algorithm

If a LMR detects that congestion occurs due to ADD conducted by the LMR at oif Ij, the LMR increases add_intvl j as follows:

Otherwise, the LMR decrease add_intvl j as follows:, >1, increase constant

, <1, increase constant

19

Extend mechanism(Cont.) A LMR recognizes that ADD causes congestion whe

n one of the following conditions is satisfied The LMR detect congestion at Ij within detect_period The LMR receives DROP_REQ(Si, Lcur(Ij, Si)) at Ij within det

ect_period (congestion at a downstream link) The LMR does not receive SESS() after detect_period has

passed since the last ADD (congestion at a upstream link)

20

Simulations Simulation parameters and

topology

The maximum number of layer : 5

The rate of each layer is CBR and 100, 100, 200, 400, 800[Kbps] from the base layer respectively

21

Simulation results

S1

LMR1

R1

LMR2

B = 1.5 Mbps

Since B is not enough for 5 layers, LMR1 increases add_intvl j gradually up to add_intvl_max to decrease the frequency of rate oscillation

23

Simulation results(Cont.)

S1

LMR1

R1

LMR2

B = 1.6 Mbps

Joins session S1

Generate CBR cross traffic with 1.3 Mbps

24

Simulation results(Cont.)

Join S1 Join S2

S1

LMR1

R1

LMR2

B = 1.7 Mbps

25

Simulation results(Cont.)

B = 1.0*N [Mbps]

(N)

Maximum Loss Rate

26

Conclusion NLM can avoid packet losses on a

lower layer and achieve fairness Simulation results presented in this

paper indicate that good responsiveness, fairness and stability of throughput of NLM compared with a receiver-driven scheme