Performance Analysis for VoIP System B92902088 邱柏儒 B92902093 紀忠毅 B92902106 莊典融 B92902120 孟昭宏 B92902088 邱柏儒 B92902093 紀忠毅 B92902106 莊典融 B92902120

PerformanceAnalysis for VoIP SystemPerformanceAnalysis for VoIP System

B92902088 邱柏儒B92902093 紀忠毅B92902106 莊典融B92902120 孟昭宏

B92902088 邱柏儒B92902093 紀忠毅B92902106 莊典融B92902120 孟昭宏

AgendaAgenda

Modeling VoIP (莊典融 )

VoIP in Ethernet (紀忠毅 )

An Example in Performance AnalysisVoIP in Wireless LAN (孟昭宏 )

Solutions to Performance Problems in VoIP over a 802.11 Wireless LAN

Summary (邱柏儒 )

Modeling VoIP (莊典融 )

VoIP in Ethernet (紀忠毅 )

An Example in Performance AnalysisVoIP in Wireless LAN (孟昭宏 )


Summary (邱柏儒 )

AgendaAgenda

Modeling VoIPVoIP in Ethernet

An Example in Performance AnalysisVoIP in Wireless LAN


Summary


An Example in Performance AnalysisVoIP in Wireless LAN


Summary

Modeling VoIPModeling VoIP

How to model VoIP trafficModeling by distributionModeling by state diagram

Pros and Cons

How to model VoIP trafficModeling by distributionModeling by state diagram

Pros and Cons

Modeling by Distribution(1)Modeling by

Distribution(1)Modeling data traffic

The data size distribution of Internet traffic which uses the TCP (many smaller files, few larger ones) seen as approximately Pareto-distributed.

Modeling data trafficThe data size distribution of Internet

traffic which uses the TCP (many smaller files, few larger ones) seen as approximately Pareto-distributed.

)2()1(,

1

,)(

2

22

1

aa

ab

a

ab

bxx

abxP

a

a


Distribution(2)

Model Phase Mean Comment

Web traffic generator

1.21.52

12(kB)0.5(sec)50,10(sec)

“ object size”“inter-object”“inter-page”

HTTP reply traces 1.04-1.14

FTP traffic 1.18 80(kB) Exponential session and burse inter-arrival time

Published Pareto distribution parameters used in modeling internet data traffic. Choose appreciate environment

Modeling data trafficModeling data traffic


Distribution(3)Run SimulatorAnalysis and diagnosis

DelayPacket lossJitters

Run SimulatorAnalysis and diagnosis

DelayPacket lossJitters

Modeling by State Diagram(1)


Modeling speech processModeling speech process



Modeling speech processModeling speech process

Pros and ConsPros and Cons

Inaccurate on modelingVariant and complex

Simulator is different from real worldUnexpected problems on hardware

Low costWe can implement solution after

getting good simulation.

Inaccurate on modelingVariant and complex

Simulator is different from real worldUnexpected problems on hardware

Low costWe can implement solution after

getting good simulation.

AgendaAgenda


An Example in Performance AnalysisVoIP in Wireless Lan

Solutions to Performance Problems in VoIP over a 802.11 Wireless Lan

Summary


An Example in Performance AnalysisVoIP in Wireless Lan

Solutions to Performance Problems in VoIP over a 802.11 Wireless Lan

Summary

Critical VoIP Performance Challenges

Critical VoIP Performance Challenges

Latency Jitter Packet

Loss Echo

Latency Jitter Packet

Loss Echo Receiver

(Packet Lost)

1 2 3 4 5

Packet Interval

Sender

Receiver (Jitter)

1 3 5

1 2 3 4 5

Receiver (Delay)

1 2 3 4

LatencyLatency

1. Good: < 80msAcceptable: 150~180ms (each way)

2. Must be addressed with VoIP protocols. Eg, SIP, H.323

3. Commonly associated with network congestion and poor bandwidth management. Not in LAN but at LAN/WAN boundary.

1. Good: < 80msAcceptable: 150~180ms (each way)

2. Must be addressed with VoIP protocols. Eg, SIP, H.323

3. Commonly associated with network congestion and poor bandwidth management. Not in LAN but at LAN/WAN boundary.

LatencyLatency

4. Minimize delay/latency Queuing techniques

Eq, DiffServ, 802.1p/q Voice packet priority over other traffic

More stringent, intelligent bandwidth management/QoS Guaranteed amount of bandwidth to

each traffic type.

4. Minimize delay/latency Queuing techniques

Eq, DiffServ, 802.1p/q Voice packet priority over other traffic

More stringent, intelligent bandwidth management/QoS Guaranteed amount of bandwidth to

each traffic type.

JitterJitter

1. Tolerance range: 20~30ms2. Possible solutions

Jitter buffer Temporarily store Smooth out the delivery of voice packet

Router queue

1. Tolerance range: 20~30ms2. Possible solutions

Jitter buffer Temporarily store Smooth out the delivery of voice packet

Router queue

JitterJitter

3. Prevent jitter TCP rate control (for data traffic) UDP rate control (for voice traffic)

Eq, Packeteer’s Application Traffic Management System

Policy-based bandwidth management or QoS strategy

3. Prevent jitter TCP rate control (for data traffic) UDP rate control (for voice traffic)

Eq, Packeteer’s Application Traffic Management System

Policy-based bandwidth management or QoS strategy

Packet LossPacket Loss

1. Loss rate < 1%: OK2. Loss rate > 3%: conversation

seems “breaking up”3. IP: best effort4. Serious packet loss may cause

dropped calls or even system failure

1. Loss rate < 1%: OK2. Loss rate > 3%: conversation

seems “breaking up”3. IP: best effort4. Serious packet loss may cause

dropped calls or even system failure

Packet LossPacket Loss

5. Prevent packet loss Apply more control IP: best effort predictable

5. Prevent packet loss Apply more control IP: best effort predictable

Solution to Improve Quality

Solution to Improve Quality

• PLC (Packet Lost Concealment) Packet Lost

• Dynamic Jitter Buffer Jitter

• Bandwidth Reservation / Packet Priorities / Queuing Delay

• G.168 Echo cancellation Echo

• VAD (Voice Active Detection) Save Bandwidth

• PLC (Packet Lost Concealment) Packet Lost

• Dynamic Jitter Buffer Jitter

• Bandwidth Reservation / Packet Priorities / Queuing Delay

• G.168 Echo cancellation Echo

• VAD (Voice Active Detection) Save Bandwidth

An Example in Performance Analysis

An Example in Performance Analysis

Testing coverageTesting environmentTesting Equipment & SoftwareREDCOM performer

QProMediaPro

Testing coverageTesting environmentTesting Equipment & SoftwareREDCOM performer

QProMediaPro

Testing CoverageTesting Coverage

1. Functionality e.g VoIP/PSTN call, QoS, SIP/Phone Setting

2. Performance e.g QoS/RTP measurement

3. Stress e.g VoIP call with data integration

1. Functionality e.g VoIP/PSTN call, QoS, SIP/Phone Setting

2. Performance e.g QoS/RTP measurement

3. Stress e.g VoIP call with data integration

Testing CoverageTesting Coverage

4. Reliability e.g Long term VoIP / Continuous VoIP call

5. Interoperability e.g Cisco ATA/IP Phone/SoftPhone

4. Reliability e.g Long term VoIP / Continuous VoIP call

5. Interoperability e.g Cisco ATA/IP Phone/SoftPhone

Testing Environment (1/2)Testing Environment (1/2)

Testing Environment (2/2)Testing Environment (2/2)

Pure EnvironmentDirect connection

Congestion EnvironmentSmart-bit tools

Pure EnvironmentDirect connection

Congestion EnvironmentSmart-bit tools

Testing Equipment & Software

Testing Equipment & Software

EXCEL 9000

• PSTN simulator

RADCOM Performer

• QPro –Voice quality measurement

• MediaPro – VoIP protocol analysis

ProLAB

• SIP Proxy server / H.323 Gateway

• SIP UA simulator / H.323 Client

VoIP Phone

• Cisco 7940 IP Phone

• XTEN / Windows Messenger

Analysis Tool

Ethereal / CoolEdit

REDCOM PerformerREDCOM Performer

QProProvide voice quality measurement

MediaProProvide VoIP protocol flow analysis

QProProvide voice quality measurement

MediaProProvide VoIP protocol flow analysis

QPro – Line ConfigurationQPro – Line Configuration

QPro – Phone ConfigurationQPro – Phone Configuration

QPro – Call SettingQPro – Call Setting

QPro – Test Result (1/2)QPro – Test Result (1/2)

QPro – Test Result (2/2)QPro – Test Result (2/2)

QPro - SummaryQPro - Summary

AgendaAgenda


A Case Study in Performance AnalysisVoIP in Wireless LAN


Summary




Summary

VoIP in Wireless LAN - outline


IntroductionVoIP Multiplex Multicast

SchemeCapacity AnalysisConclusion



The Problems faced by WLAN

The Problems faced by WLAN

System CapacitySystem capacity for voice can be quite low

Other data trafficData from traditional App can interfere with each other

System CapacitySystem capacity for voice can be quite low

Other data trafficData from traditional App can interfere with each other

VoIP in 802.11bVoIP in 802.11b

VoIP in WLAN can potentially support more than 500 sessions in theory

In practice, only 12 are supported due to various overhead

VoIP in WLAN can potentially support more than 500 sessions in theory

In practice, only 12 are supported due to various overhead


Support data rate up to 11Mb/sA VoIP stream typically requires

less than 10kb/sThe number of simultaneous VoIP

streams that can be supported by an 802.11b in theory is around 11M/10K = 1100

About 550 VoIP sessions

Support data rate up to 11Mb/sA VoIP stream typically requires

less than 10kb/sThe number of simultaneous VoIP

streams that can be supported by an 802.11b in theory is around 11M/10K = 1100

About 550 VoIP sessions


In practice, no more than a few VoIP sessions

If GSM 6.10 codec is used, the maximum is 12

The result is mainly due to added packet header overheads as well as the inefficiency inherent in the WLAN MAC

In practice, no more than a few VoIP sessions

If GSM 6.10 codec is used, the maximum is 12

The result is mainly due to added packet header overheads as well as the inefficiency inherent in the WLAN MAC


IP + UDP + RTP header = 40bytesVoIP payload ranging from 10 to

30 bytesThe transmission time:

30 * 8 / 11 = 22 us40 * 8 / 11 = 29 us

Efficiency drops to less than 50%

IP + UDP + RTP header = 40bytesVoIP payload ranging from 10 to

30 bytesThe transmission time:

30 * 8 / 11 = 22 us40 * 8 / 11 = 29 us

Efficiency drops to less than 50%


Physical layer have additional overhead more than 800 us

Attributed to the Physical preamble, MAC header, MAC backoff time, MAC ACK, Inter-transmission time

Overall efficiency drops to less than 3%

Physical layer have additional overhead more than 800 us

Attributed to the Physical preamble, MAC header, MAC backoff time, MAC ACK, Inter-transmission time

Overall efficiency drops to less than 3%


TCP connection will cause unacceptably large increase in the delay and packet loss rate of VoIP traffic

TCP connection will cause unacceptably large increase in the delay and packet loss rate of VoIP traffic







Multiplex-Multicast Scheme


An 802.11 WLAN is referred to as the basic service set (BSS) in the standard specification

There are two types of BSSs: Independent BSS and

Infrastructure BSS.

An 802.11 WLAN is referred to as the basic service set (BSS) in the standard specification

There are two types of BSSs: Independent BSS and

Infrastructure BSS.

Ad Hoc (Independent) BSSAd Hoc (Independent) BSS

Infrastructure BSSInfrastructure BSS



Focus on infrastructure BSSAssume that all voice streams are

between stations in different BSSEach AP has two interfaces, an

802.11 interface, and an Ethernet interface which is connected to the voice gateway.

Focus on infrastructure BSSAssume that all voice streams are

between stations in different BSSEach AP has two interfaces, an

802.11 interface, and an Ethernet interface which is connected to the voice gateway.





Within a BSS, there are two streams for each VoIP session.

M-M Scheme idea : Combine the data from several

downlink streams into a single packet for multicast over the WLAN to their destinations

Within a BSS, there are two streams for each VoIP session.

M-M Scheme idea : Combine the data from several

downlink streams into a single packet for multicast over the WLAN to their destinations





The voice multiplexer resides in the voice gateway for H.323

The MUX can also resides in a specially designed AP or a server between the voice gateway and AP

The voice multiplexer resides in the voice gateway for H.323

The MUX can also resides in a specially designed AP or a server between the voice gateway and AP

Multiplex-Multicast Procedure


The download link VoIP traffic first goes through a MUX in the voice gateway

The MUX replaces the RTP, UDP, IP header of each packet with a compressed mini header

In mini header, there is an ID used to identify the session of the VoIP packet

The download link VoIP traffic first goes through a MUX in the voice gateway

The MUX replaces the RTP, UDP, IP header of each packet with a compressed mini header

In mini header, there is an ID used to identify the session of the VoIP packet



Then multicast the multiplexed packet to the WLAN through AP

The DEMUX at the receiver restores the original RTP header and necessary information

Then multicast the multiplexed packet to the WLAN through AP

The DEMUX at the receiver restores the original RTP header and necessary information



Header

Data1

Header

Header

Data2

Data3

HeaderMinih + Data1 + Minih + Data2

…



AdvantagesAdvantages

Reduce the number of VoIP streams in one BSS from 2n to 1 + n, where n is the number of VoIP sessions.

Improve the bandwidth efficiency

Reduce the number of VoIP streams in one BSS from 2n to 1 + n, where n is the number of VoIP sessions.

Improve the bandwidth efficiency

TradeoffTradeoff

The MUX sends out a multiplexed packet every T ms, which is equal to or shorter than the VoIP inter-packet interval.

For GSM 6.10, the inter-packet interval is 20 ms.

Larger value of T can improve bandwidth efficiency

The MUX sends out a multiplexed packet every T ms, which is equal to or shorter than the VoIP inter-packet interval.

For GSM 6.10, the inter-packet interval is 20 ms.

Larger value of T can improve bandwidth efficiency

Tradeoff (cont.)Tradeoff (cont.)

But the larger T will cause more delay

One can control the tradeoff between bandwidth efficiency and delay by adjusting T

But the larger T will cause more delay

One can control the tradeoff between bandwidth efficiency and delay by adjusting T





Should be solved by encrypting the voice packet

Should be solved by encrypting the voice packet







CSMA/CA AlgorithmCSMA/CA Algorithm

Capacity AnalysisCapacity Analysis

n : maximum number of sessions that can be supported

Tdown & Tup: transmission times for downlink and uplink packets

Tavg: average time between the transmissions of two consecutive packets in a WLAN

NP : number of packets sent by one stream in one second

1/Tavg = number of streams * NP

n : maximum number of sessions that can be supported

Tdown & Tup: transmission times for downlink and uplink packets

Tavg: average time between the transmissions of two consecutive packets in a WLAN

NP : number of packets sent by one stream in one second

1/Tavg = number of streams * NP

Capacity of Ordinary VoIPCapacity of Ordinary VoIP

OHhdr = HRTP + HUDP + HIP + HMAC

OHsender = DIFS + averageCW + PHY if unicast packet:

OHreceiver = SIFS + ACK

Tdown = Tup = (Payload +OHhdr) * 8 / dataRate + OHsender + OHreceiver

OHhdr = HRTP + HUDP + HIP + HMAC

OHsender = DIFS + averageCW + PHY if unicast packet:

OHreceiver = SIFS + ACK

Tdown = Tup = (Payload +OHhdr) * 8 / dataRate + OHsender + OHreceiver

Capacity of Ordinary VoIPCapacity of Ordinary VoIP

n downlink and n uplink unicast streams

Tavg = (Tdown + Tup) / 2

1/Tavg = 2n *Np

n = 11

n downlink and n uplink unicast streams

Tavg = (Tdown + Tup) / 2

1/Tavg = 2n *Np

n = 11

Capacity of M-M SchemeCapacity of M-M Scheme

the RTP, UDP and IP header of each packet is compressed to 2 bytes

Tdown = [(Payload + 2) *n + HUDP + HIP + HMAC] * 8 / dataRate + OHsender

Tavg = (Tdown + n *Tup) / (n + 1)

1/Tavg = (n + 1) *Np n = 21.2

the RTP, UDP and IP header of each packet is compressed to 2 bytes

Tdown = [(Payload + 2) *n + HUDP + HIP + HMAC] * 8 / dataRate + OHsender

Tavg = (Tdown + n *Tup) / (n + 1)

1/Tavg = (n + 1) *Np n = 21.2

Capacities assuming Different Codecs

Capacities assuming Different Codecs

Codecs Ordinary VoIPMultiplex-Multicast

Scheme

GSM 6.10 11.2 21.2

G.711 10.2 17.7

G.723.1 17.2 33.2

G.726-32 10.8 19.8

G.729 11.4 21.7

SimulationsSimulations

increase the number of VoIP sessions until the per stream packet loss rate exceeds 1%

system capacity = max number of sessions

assume that the retry limit for each packet is 3

increase the number of VoIP sessions until the per stream packet loss rate exceeds 1%

system capacity = max number of sessions

assume that the retry limit for each packet is 3

SimulationsSimulations

For ordinary VoIP over WLAN, the system capacity is 12

Exceeding the system capacity leads to a large surge in packet losses for the downlink streams

For ordinary VoIP over WLAN, the system capacity is 12

Exceeding the system capacity leads to a large surge in packet losses for the downlink streams

Analysis V.S. SimulationAnalysis V.S. Simulation

Different Schemes Analysis Simulation

Original VoIP 11.2 12

Multiplex-MulticastScheme

21.2 22







ConclusionsConclusions

M-M scheme can reduce the large overhead when VoIP traffic is delivered over WLAN

it requires no changes to the MAC protocol at the wireless end stations

more readily deployable over the existing network infrastructure.

it makes the voice capacity nearly 100% higher than ordinary VoIP

M-M scheme can reduce the large overhead when VoIP traffic is delivered over WLAN

it requires no changes to the MAC protocol at the wireless end stations

more readily deployable over the existing network infrastructure.

it makes the voice capacity nearly 100% higher than ordinary VoIP

AgendaAgenda




Summary




Summary

Summary(1/2)Summary(1/2)

Mathematical modeling- Queuing theory Poisson distribution Pareto distribution- Inaccuracy

Industrial testing method Performance in various

environments- Ethernet- Wireless

Mathematical modeling- Queuing theory Poisson distribution Pareto distribution- Inaccuracy

Industrial testing method Performance in various

environments- Ethernet- Wireless

Summary(2/2)Summary(2/2)

Improvement- e.g. priority queue, jitter absorption, loss concealment. - QoS absolute QoS level relative QoS level e.g. Expedited Forwarding, Assured Forwarding.

Improvement- e.g. priority queue, jitter absorption, loss concealment. - QoS absolute QoS level relative QoS level e.g. Expedited Forwarding, Assured Forwarding.

ReferenceReference

P.M. Fiorini: Voice over IP for Enterprise Networks: Performance Implications & Traffic Models

E. Noel & K.W. Tang: Performance Analysis of a VoIP Access Architecture

Wei Wang: Solutions To Performance Problems In VoIP Over A 802.11 Wireless LAN

P.M. Fiorini: Voice over IP for Enterprise Networks: Performance Implications & Traffic Models

E. Noel & K.W. Tang: Performance Analysis of a VoIP Access Architecture

Wei Wang: Solutions To Performance Problems In VoIP Over A 802.11 Wireless LAN

Q & AQ & A

Thanks for your listening…

Thanks for your listening…

Documents

Performance Analysis for VoIP System B92902088 邱柏儒 B92902093 紀忠毅 B92902106 莊典融 B92902120 孟昭宏 B92902088 邱柏儒 B92902093 紀忠毅 B92902106 莊典融 B92902120