Experimental Experimental Measurement of VoIP Measurement of VoIP Capacity in IEEE 802.11 Capacity in IEEE 802.11 WLANsWLANs

Sangho ShinHenning Schulzrinne

Department of Computer ScienceColumbia University

VoIP over Wireless LANsVoIP over Wireless LANs

InternetInternet

AP (Access Point)

PBX

WIFI

Motivation and goalMotivation and goal

Check the VoIP capacity using wireless cards and compare it with theoretical and simulation results

Identify all factors that affect the VoIP capacity in experiments and simulations

OutlineOutline

Theoretical capacity for VoIP traffic VoIP capacity via simulations VoIP capacity via experiments ‘Hidden factors’ that affect

experiments and simulations Conclusion

Packetization interval

1 2 3 N 1 2 3 N……. …….MAC

Theoretical capacityTheoretical capacity

parameters value

Voice codec 64 kb/s

Packet size 160B

Packetization interval

20ms

Transport layer UDP

PHY data rate 11 Mb/s

RTS/CTS No

bt TT

PN

2max

Capacity (calls)

Packetization Interval (ms)

= 15 calls

PLCP = Physical Layer Convergence Procedure

PLCP MAC IP UDP Voice ACKPLCPbackoff

DIFS SIFS

Tt

Tb

RTP

Simulation setupSimulation setup

WIFIWIFI

WIFI

WIFI

Ethernet-Wireless

parameters value

Voice codecG.711 (64 kb/s)

Packet size 160B

Packetization interval

20ms

Transport layer UDP

PHY data rate 11Mb/s

RTS/CTS No

WIFI

IEEE 802.11b

QualNet simulator v3.9

Simulation resultsSimulation results

CapacityNumber of VoIP sources

90th

per

cent

ile d

ela

y (m

s)

Downlink delay

Uplink delay

ExperimentsExperimentsNJ Rutgers University

ExperimentsExperiments

80 ft

70 ft

Atheros

Intel

Experimental setupExperimental setup

parameters value

Voice codecG.711 (64 kb/s)

Packet size 160B

Packetization interval

20ms

Transport layer UDP

PHY data rate 11Mb/s

RTS/CTS No

client

client clientclient client

clientclientclient

clients clientAPclient

client clientclientclient

IEEE 802.11bAtheros chipsetMadWifi-0.9.3

Experimental resultsExperimental results

Capacity

90th

per

cent

ile d

ela

y (m

s)

Downlink delay

Uplink delay

FactorsFactors ARF (Auto Rate Fallback) Preamble size PHY data rate of ACK frames Offset of VoIP traffic start time Signal strength Scanning APs Retry limit Network buffer size

90th

per

cent

ile d

ela

y (m

s)

Fixed rate

ARF (AMRR)

Threshold for capacity

ARFARF ARF (Auto Rate Fallback)

PHY data rate are automatically changes When frame loss is caused by bad link quality, it helps When frame loss is caused by congestion, it makes worse

Problems The effect varies according to algorithms

Turned off in simulations Turned on in wireless cards

Experimental results 8% of frames were transmitted with lower rates

AMRR=Adaptive Multi-Rate Retry

Preamble sizePreamble size IEEE 802.11b : long and short

preamble QualNet, NS-2 Long preamble Atheros + MadWifi driver Short

preamble Theoretical capacity with the long

preamble = 12 calls Experimental results

Long Short

Preamble size 144 us

72 us

Header size (us) 48 us 24 us

Total size (us) 192 us

96 us

Fraction in a VoIP (size)

9% 6%

Fraction in a VoIP (time)

53% 36%

PLCP = Physical Layer Convergence Procedure

90th

per

cent

ile d

ela

y (m

s)

Short

Long

PHY data rate for ACK PHY data rate for ACK framesframes ACK frames

Required for ARQ Theoretical VoIP capacity

using 11 Mb/s for ACK frames 16 calls

Experimental results

PLCP MAC

14B

2Mb/s 152 us = 57% of a VoIP packet11Mb/s106 us = 39% of a VoIP packet

Type : 01 Subtype 1101

90th

per

cent

ile d

ela

y (m

s)

11 Mb/s

2 Mb/s

MadWifi2Mb/sQualNet11Mb/sNS-21Mb/s

Offset of VoIP traffic start Offset of VoIP traffic start timetime

1 2 3 4

Packetization interval

1 2 3 4Application layerOffset

MAC layer data backoff

SIFS

ACK

DIFS

data

VoIP source 1

VoIP source 2

VoIP source 3

VoIP source 4

1

2

3

4

1

2

3

4

MAC layer 1 2 3 4 1 2 3 4collisions

Offset of VoIP traffic start Offset of VoIP traffic start timetime

Uplink retry rate

650 μs = the optimal offset (20ms/(15 sources*2))

Offset of traffic start time (μs)

Simulation results with 15 VoIP sources

90th

per

cent

ile d

ela

y (m

s)

FactorsFactors ARF (Auto Rate Fallback) Preamble size PHY data rate of ACK frames Offset of VoIP traffic start time Signal strength Scanning APs Retry limit Network buffer size

Fixed

Short

2Mb/s

Randomized

Signal strengthSignal strength

Scanning APsScanning APs Scan APs based on

signal strength transmission failure Regularly (e.g. every min) Hard to determine the

algorithms Problems

Management frames have a higher priority than data frames causes delay

Increases the traffic make channels congested

1 probe request and 1 ~ 2 probe responses per channel

APclientProbe request (broadcast)

Probe response (unicast)

(fo

r 10

0 s)

Retry limitRetry limit Wireless nodes retransmit frames until the number

of retransmission reaches the retry limit Long retry limit - frame size > RTS threshold Short retry limit - frame size ≤ RTS threshold

Effect More retransmissions reduces packet loss, but

increases congestion Less retransmissions Increases the packet loss

Experimental results

(4)(7)

Network buffer sizeNetwork buffer size Packet loss happens mostly because of the buffer

overflow at the AP Small buffer increase the packet loss Bigger buffer reduces packet loss, but increase the

delay Buffer size needs to be big enough to allow 60ms of

delay Simple static queuing analysis

avgS

BD

1

max

Maximum queuing delay

Buffer size

Packet size

Average service rate

µ = 1/500D = 60msS = 200BBmin = 5.8KB < 10KB MadWifi

ConclusionConclusion Need to consider the following factors when

measuring the VoIP capacity experimentally ARF Preamble size PHY data rate of ACK frames Offset of VoIP traffic start time Scanning APs Retry limit Network buffer size

By adjusting all the factors, we can achieve the same experimental, simulation, theoretical capacity

Our study can be used in any 802.11 experiments and the analysis and comparison

Thank you!Thank you!