Evaluation of Multiplexing and Buffer Policies Influence on VoIP Conversation Quality

Presentación

Jose Saldana

Jenifer Murillo

Julián Fernández Navajas

José Ruiz Mas

Eduardo Viruete Navarro

José I. Aznar

O M U N I C A C I O N E S

TRUPO DE

CECNOLOGÍAS

GDE LAS

CPS - University of Zaragoza, Spain

INTRODUCTION RELATED WORKS TEST METHODOLOGY RESULTS CONCLUSIONS

Index


Index

CCNC January 9-11, 2011. Las Vegas Evaluation of Multiplexing and Buffer Policies Influence on VoIP

4

Introduction

The use of Internet for multimedia transmission is growing as bandwidth increases.

Services with hard real-time

requirements:

- VoIP: Voice over IP

- Videoconferencing

- Online Gaming


INTRODUCTION RTP COMPRESSION RTP MULTIPLEXING ROUTER R-FACTOR


5

Introduction

- The use of best-effort networks for these services is a problem for the experienced quality

- Used protocols:

- Signaling: SIP, H.323

- Media transport: RTP, simple UDP




6

Introduction

Real-time requirements force us to divide the information into small pieces, and send it using a small time period, and small packets.

This implies an overhead, as every packet needs the IP, UDP and maybe RTP headers




7

RTP packet

RTP packets’ overhead

VoIP packet with 2 G.729a samples

Efficiency: 33% for IPv4

IP header UDP header RTP header Sample Sample

8 bytes20 bytes 12 bytes 10 bytes 10 bytes




8

Scenarios where many multimedia flows share a path. Does it represent an advantage?

Office

Data centre




9

Scenarios where many multimedia flows share a path. Does it represent an advantage?

Internet café




10

Two possible improvements Improvement 1: RTP compression schemes:

- CRTP: RFC 2508, February 1999

- ECRTP: RFC 3545, July 2003: Enhanced CRTP for scenarios with packet loss, packet reordering and long delays.

- ROHCv2: RFC 5225, April 2008

They use the repeatability of IP/UDP/RTP headers to compress them.




11

RTP compression - The two extremes share a context.

- Every packet carries a CID

- Some fields are avoided, other are compressed (delta compression, etc.) or inferred from lower layers

Problems:

- Only hop-by-hop usage

- Synchronization of the context




12

Increasing the number of samples

Improvement 2: More samples in a single packet. If they belong to the same flow:




13


If different flows share the same path (voice trunking) the packet frequency can be the same. Added delays?




14





15


Maximum added delay: Packet period. Independent of the number of flows




16

IP header UDP header RTP header

IP header L2TP

Sample Sample IP header UDP header RTP header Sample Sample IP header UDP header RTP header Sample Sample

RH Sample Sample RH Sample Sample RH Sample Sample

PPP PPPMux PPPMux PPPMux

RTP multiplexing

Combines header compression and multiplexing a number of flows. Advantages:

- Reducing overhead, bandwidth saving

- Reducing packets per second

Real scale




17

RTP multiplexing

Disadvantages:

- New added delays

- Processing charge

Increasing packet size: Good or bad?

Real scale




IP header L2TP





18

Influence of the router

- The amount and size distribution of background traffic will affect the real-time traffic.

- Packet loss can be modified with the change of packet size, depending on the policy of the router’s buffer.

- Routers have a pps limitation.




19

Motivation of this work

- Study the influence of buffer policies and multiplexing schemes on the perceived quality, for real-time services.

- Service used: VoIP. Representative.

- Real-time requirements

- Wide-deployed service

- Existence of scenarios where many flows share the same path




20

R-factor - Defined by ITU G.107 (E-Model)

- Ranges from 0 (bad quality) to 100 (good)

- Acceptable for R > 70

- Dependence on delay and packet loss

- Widely accepted quality estimator for VoIP services




Index


22

RTP multiplexing proposals - Different proposals. We will consider:

- TCRTP (RFC 4170): Tunneling Multiplexed Compressed RTP.

- Sze: «A Multiplexing Scheme for H.323 VoIP Applications», IEEE J. Selected Areas Comm., Sep 2002.


MULTIPLEXING PROPOSALS MULTIPLEXING USES BUFFER POLICIES


23

TCRTP - It does not define a new protocol.

Combines some of them.


PPP

PPP Mux

ECRTP

samples

IP

UDP

RTP...

ECRTP

samples

L2TP

IP



24

- Different Reduced Header sizes (ECRTP)

- The use of the L2TP tunnel makes it possible to use ECRTP end-to-end

Example: 3 RTP packets with 2 samples per packet:

IP header L2TP RH Sample Sample RH Sample Sample RH Sample Sample


TCRTP




25

Sze - Includes many RTP packets into a UDP one

- Different Reduced Header sizes

- Need of some tables at the origin and destination. Non standard

Exampe: 3 RTP packets with 2 samples per packet:



IP header UDP headerR

HSample Sample

R

HSample Sample RH Sample Sample


26

Packet size comparative 3 multiplexed packets

5 multiplexed packets



Real scale

RTP

TCRTP

Sze


IP header L2TP

IP header UDP header



R

HSample Sample

R




IP header L2TP

IP header UDP header

Sample Sample IP header UDP header RTP header Sample Sample IP header UDP header RTP header Sample Sample IP header UDP header RTP header Sample Sample IP header UDP header RTP header Sample Sample

RH Sample Sample RH Sample Sample RH Sample Sample RH Sample Sample RH Sample Sample

R

HSample Sample

R

HSample Sample

R

HSample Sample

R


PPP PPPMux PPPMux PPPMux PPPMuxPPPMux

RTP

TCRTP

Sze


27

RTP multiplexing uses - Bandwidth relationship

- Packets per second: reduced by k



0,3

0,4

0,5

0,6

0,7

0,8

0,9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

X

k

Bandwidth Saving X for p = 0.95 S=10 bytes

Xrh S=10

S=20 bytes

Xrh=20

S=30 bytes

Xrh S=30

BW compressed/BW native


28

RTP multiplexing uses - Packet size increase



0

100

200

300

400

500

600

700

800

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

byte

s

k

Packet sizeRTP S=10 bytes

RTP S=20 bytes

RTP S=30 bytes

TCRTP S=10 bytes

TCRTP S=20 bytes

TCRTP S=30bytes


29

- «Rule of the thumb»: Bandwidth-delay product.

- «Stanford model»: Division by sqrt(N) (N:number of TCP flows).

- Other proposal: time-limited buffer. Interesting for this work. Limits OWD. But penalizes big packets

Buffer size and buffer policies




30

Buffer size and buffer policies - Multiplexing tradeoffs



Bandwidth saving

Multiplexing

Bigger packet size

Packet loss increase

Packet loss reduction

Buffer policy

Added delays R-factor

R-factor

R-factor

Background traffic

Reduced pps Packet loss reduction

R-factor

Router limitation


31

Buffer size and buffer policies - We will compare

- Dedicated buffer: Only VoIP

- High-capacity buffer

- Time-limited buffer



IP network

MUX DEMUX .

.

.

.

.

.

RTP RTP multiplexing RTP


Index


33

General Scheme - Use of a testbed

- Generator: D-ITG: Statistics of packet size and inter packet delay.


GENERAL SCHEME TRAFFIC GENERATION SYSTEM DELAYS

Traffic

Generation

RouterTraffic

Capture

Real Traffic in a testbed

Network

delays

+

Dejitter

buffer

Offline post-processing

Traffic

Trace

Final

Results

VoIP

Background

Buffer

policies


34

Traffic generation - Background traffic

- 50% 40 bytes

- 10% 576 bytes

- 40% 1500 bytes

- Only UDP, in order to avoid flow control: always the same background traffic.

- Different rates to saturate the access router

- Network does not loose packets




35

Traffic generation - Multiplexed VoIP traffic

- Three header sizes:

- COMPRESSED_RTP 97.3%

- COMPRESSED_UDP 2.6%

- FULL_HEADER 0.0033% (negligible)

- Packet size: Binomial k, p=0.973

- Used for TCRTP and Sze.

- 400 seconds of traffic. First and last 20 are discarded




36

System delays - Packetization delay: G.729a.

- 15, 25 or 35 ms

- Retention time at the mux: Packet period

- Process: 5ms


IP network

MUX DEMUX

Tprocess Tqueue Tnetwork Tprocess TdejitterTretentionTpacketization

.

.

.

.

.

.




37


- 15, 25 or 35 ms


- Process: 5ms


IP network

MUX DEMUX


.

.

.

.

.

.




38


- 15, 25 or 35 ms


- Process: 5ms


IP network

MUX DEMUX


.

.

.

.

.

.




39

System delays (2) - Queuing delay

- Network delay

- Fixed: 20ms

- Lognormal: avg 20ms. Variance 5


IP network

MUX DEMUX


.

.

.

.

.

.




40

System delays (2) - Queuing delay

- Network delay

- Fixed: 20ms

- Lognormal: avg 20ms. Variance 5


IP network

MUX DEMUX


.

.

.

.

.

.




41

System delays (3) - De-jitter buffer: adds delays and packet

losses:

- loss de-jitter buffer ~ P { l > bg}

- delay de-jitter buffer : number of samples

- Buffer size: maximize R-factor. Static.


IP network

MUX DEMUX


.

.

.

.

.

.




Index


43

Dedicated buffer - 200 kbps only for VoIP


DEDICATED BUFFER HIGH CAPACITY BUFFER TIME-LIMITED BUFFER

76

77

78

79

80

81

82

5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

R-f

ac

tor

Number of multiplexed calls k

R-factor

RTP

TCRTP

Sze


44

Dedicated buffer - 200 kbps only for VoIP



76

77

78

79

80

81

82

5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

R-f

ac

tor

Number of multiplexed calls k

R-factor

RTP

TCRTP

Sze

Effect of bandwidth saving

Effect of added delays


45

High capacity buffer - 1Mbps shared. 5 flows


60

65

70

75

80

85

400 450 500 550 600 650 700 750 800 850 900 950 1000

R-f

acto

r

background traffic (kbps)

R factor 1 RTP

5 RTP

5 TCRTP

5 Sze



46



60

65

70

75

80

85

400 450 500 550 600 650 700 750 800 850 900 950 1000

R-f

acto

r


R factor 1 RTP

5 RTP

5 TCRTP

5 Sze


Step-like graphs. When the bandwidth is not enough, the quality falls


47



60

65

70

75

80

85

400 450 500 550 600 650 700 750 800 850 900 950 1000

R-f

acto

r


R-factor 1 RTP

10 RTP

10 TCRTP

10 Sze



48



60

65

70

75

80

85

400 450 500 550 600 650 700 750 800 850 900 950 1000

R


R factor 1 RTP

15 RTP

15 TCRTP

15 Sze



49



60

65

70

75

80

85

400 450 500 550 600 650 700 750 800 850 900 950 1000

R


R factor 1 RTP20 RTP20 TCRTP20 Sze


The bigger the number of flows, the bigger the bandwidth saving


50

Time-limited buffer - 1Mbps shared. 5 flows


66

68

70

72

74

76

78

80

82

400 450 500 550 600 650 700 750 800 850 900 950 1000

R-f

acto

r


R-factor1 RTP

5 RTP

5 TCRTP

5 Sze



51



66

68

70

72

74

76

78

80

82

400 450 500 550 600 650 700 750 800 850 900 950 1000

R-f

acto

r


R-factor1 RTP

5 RTP

5 TCRTP

5 Sze


Time-limited buffer: slope instead of step-like


52



66

68

70

72

74

76

78

80

82

400 450 500 550 600 650 700 750 800 850 900 950 1000

R-f

acto

r


R-factor1 RTP

5 RTP

5 TCRTP

5 Sze


Native RTP behaves better than multiplexing schemes


53



66

68

70

72

74

76

78

80

82

400 450 500 550 600 650 700 750 800 850 900 950 1000

R-f

acto

r


R-factor1 RTP

10 RTP

10 TCRTP

10 Sze



54



66

68

70

72

74

76

78

80

82

400 450 500 550 600 650 700 750 800 850 900 950 1000

R-f

acto

r


R-factor1 RTP

10 RTP

10 TCRTP

10 Sze


Gaining zone

0.5% worse


55

Time-limited buffer


0

5

10

15

20

400 450 500 550 600 650 700 750 800 850 900 950 1000

Pac

ket

Loss

(%

)


Background Traffic Packet Loss 1 RTP

10 RTP

10 TCRTP

10 Sze


The bigger the bandwidth saving, the smaller the packet loss


56



60

62

64

66

68

70

72

74

76

78

80

82

400 450 500 550 600 650 700 750 800 850 900 950 1000

R-f

acto

r


R-factor

1 RTP

15 RTP

15 TCRTP

15 Sze



57



60

62

64

66

68

70

72

74

76

78

80

82

400 450 500 550 600 650 700 750 800 850 900 950 1000

R-f

acto

r


R-factor

1 RTP

20 RTP

20 TCRTP

20 Sze



58

Time-limited buffer


-10

-5

0

5

10

15

20

25

400 450 500 550 600 650 700 750 800 850 900 950 1000

% R

-fac

tor

imp

rove

me

nt


% R-factor improvement5 TCRTP5 Sze10 TCRTP10 Sze15 TCRTP15 Sze20 TCRTP20 Sze



Index


60

Conclusions - Comparison of multiplexing schemes

- Service used: VoIP

- New delays added: small. The bandwidth saving is significant

- R-factor

- improvement up to 20%

- impairment 1%

- Packet size is important, depending on buffer policies



61

Conclusions - Is it better to use only one tunnel or to

group calls into a number of tunnels?


50

55

60

65

70

75

80

85

400 450 500 550 600 650 700 750 800 850 900 950 1000

R


R factor

20 RTP

20 TCRTP

2x10 TCRTP


62

Conclusions - Is it better to use only one tunnel or to

group calls into a number of tunnels?


50

55

60

65

70

75

80

85

400 450 500 550 600 650 700 750 800 850 900 950 1000

R


R factor

20 RTP

20 TCRTP

2x10 TCRTP

Presentación

Jose Saldana

Jenifer Murillo

Julián Fernández Navajas

José Ruiz Mas

Eduardo Viruete Navarro

José I. Aznar

O M U N I C A C I O N E S

TRUPO DE

CECNOLOGÍAS

GDE LAS

CPS - University of Zaragoza, Spain

Evaluation of Multiplexing and Buffer Policies Influence on VoIP Conversation Quality

Education

buffer policies influence

packet period

packet loss

packet frequency

single packet

number of samples improvement

context ccnc

number of samples maximum