Implementation of a WAP model to evaluateCapacity in 3G radio access networks

Henrik Fållby

Outline

• Scoop of this thesis• Packet switched vs. circuit switched networks• Packet Data in GSM radio networks• Wireless Application Protocol• WAP traffic model• Petra System Simulator• Evaluation results for WAP over EGPRS• Comparing EGPRS with GPRS with WAP traffic• Comparing WAP vs. WWW over EGPRS• Summary

Scoop of the Thesis

• Development of a WAP model• Evaluation of WAP traffic over GPRS and EGPRS RAN• Comparison of WAP vs. WWW performance

Packet switched contra Circuit switched networks

• Several users can chare the samechannel (Time Slot)

• Channel only used when needed• Suitable for non real-time

applications, i.e WAP/WWW

• Only one user per channel• Channel occupied until the session

is closed• Suitable for real-time applications,

i.e. voice

Circuit SwitchedPacket Switched

Packet Data in GSM radio networks:GSM radio network

GPRS nodes

IP-BackboneNetWorkOther

PLMN External X.25 Network

ExternalIP Network

SGSN

GGSN

Base Station System

MS

Traffic and signaling information

Signaling information

BSCBTS

Switching System

HLR

AUC

EIR

GMSCMSC/VLRISDN/PSTN

Packet Data in GSM radio networks:Modulation and Coding Schemes for EGPRS and GPRS

Coding Schemes

Max no of info bits per radio block

Max data rate per TS

(kbps)

CS-1 160 8.0CS-2 240 12.0CS-3 288 14.4CS-4 400 20.0

Coding Schemes Modulation

Max no of info bits per radio block

Max data rate per TS

(kbps)MCS-1 GMSK 176 8.8MCS-2 GMSK 224 11.2MCS-3 GMSK 296 14.8MCS-4 GMSK 352 17.6MCS-5 8PSK 448 22.4MCS-6 8PSK 592 29.6MCS-7 8PSK 896 44.8MCS-8 8PSK 1088 54.4MCS-9 8PSK 1184 59.2

EGPRS Modulation and Coding SchemesGPRS Coding Schemes

2φ

1φ

GMSK

1

0

2φ

1φ

8PSK

(0,1,0)

(0,0,0)

(0,0,1)

(1,0,1)

(1,1,0)

(1,1,0)

(0,1,1)

(1,1,1)

Packet data in GSM networks:GPRS multiframe structure

User dataHeader Tail

Header User data

USF RLCHeader

RLCInformation BCS

Packet

LLC PDU

RLC/MAC blocks

Radio blocks

52 TDMA frames

Network layer

LLC layer

RLC/MAC layer

Physical layer

<1.5 kbytes

~1.6 kbytes

20-50 bytes

4 x 114 bits

B0 B1 B2 X B3 B4 B5 X B7 B8 B9X B10 B11B6 X

User data Application layer

NormalBurst

NormalBurst

NormalBurst

NormalBurst

USF RLCHeader

RLCInformation BCS

By EGPRSeffected layers

RLC

8PSK

Wireless Application Protocol:Properties of a mobile terminal

LimitedKeyboard

Small Display

NarrowbandAccess

Limited Memory andComputing Power

Wireless Application Protocol:WAP architecture

Client

WAPUser

Agents

WAPProtocol

Stack

WAPGateway/

Proxy

Encoders

ProtocolConversion

OriginServer

CGIScripts,

etc.WSP request (URL)

WSP Resp. (WAP binary WML)

HTTP request (URL)

HTTP Response (WML)

WIRELESS DOMAIN:WAP

Protocols

INTERNET DOMAIN:Internet

Protocols

WMLWMLScript

Traffic model:Building a WAP traffic model

• Measurements• Analysis• Choice of model structure• Parameterisation

Packetinterarrival time

Packetsize

Sessioninterarrival time

Sessionlength

Traffic model:Distributions and parameters for the WAP model

Packetinterarrival time

Packetsize

Sessioninterarrival time

Sessionlength

Session interarrival

time [s]

Number of Packets within

a session

Packet interarrival

time [s]

Packet size [Byte]

Distribution Exponential GeometricNegative

exponentialLognormal

Minimum value 0 0 0 0Maximum value - - - 1600

Mean 1/λ 5 13.8 471Standard deviation 1/λ - - 355

PETRA System Simulator:Overview

BSS

MS

• Multiple cell simulator• Equally sized 3-sector macro

cells• 4/12 frequency reuse• TU3 no frequency hoping• Downlink• 20 ms iterations• EGPRS MCS9 with IR• GPRS CS2 without LA• Multislot allocation not modeled

PETRA system simulator:Simulation environment

MS

ApplicationIP / X.25

SNDCP

LLC

GSM RF

RLC*

MAC

GSM RF

RLC*

MAC

L1bis

NetworkService

BSSGP

SNDCP GTPRelay

LLC

BSSGP

NetworkServiceL1bis

UDP/TCP

IP

L2

L1

GTP

UDP/TCP

IP

L2

L1

IP / X.25

BSS SGSN GGSNUm Gb Gn Gi

Relay

Header dataTraffic model Detailedsimulation *TBF setup not included

Packet Emulator Tool for Radio network Application (PETRA) used tosimulate the RAN

PETRA system simulator:Packet handling

Application layer

Network layer

LLC layer

RLC/MAC layer

User Data1500 byte

LLCheader11 -12byte

RLCheaderUSF BCS

RLCInformation20-50 byte

RLCheaderUSF BCSRLC

Information

Traffic model

Application Payload

User DataHeader User Data

IP+WDP/

TCPHeader

Data

Headers presentin PETRA

Headers notpresent inPETRA

Simulation results WAP over EGPRS:CDF of packet throughput

0 10 20 30 40 50 600

10

20

30

40

50

60

70

80

90

100

C.D

.F. [

%]

Average packet throughput per timeslot [kbps]

100 WAP users per cell350 −"− 700 −"−

( ) ( ) ( ) α=≤==�=

xXPifxFx

i 0

• The 10th percentile ofpacket throughputstates that 90% of thepackets will havethroughput equal to orhigher than the plottedvalues

• Note if assuming evenpacket distributionover cell area theprobability can beseen as cell area

CDF (cumulative distribution function)Definition:

Simulation results WAP over EGPRS:10th percentile of packet throughput for WAP

0 50 100 150 2000

5

10

15

20

25

30

35

Average offered load [kbps]

10th

per

cent

ile o

f pac

ket t

hrou

ghpu

t per

tim

eslo

t [kb

ps]

(25)

(50)

(100)

(200)

(300)(350)

(400)(500) (600)

(700)(800)

(In parenthesis average number of users per cell)

•QoS target/reference point:Definition:90% of the packets shall have athroughput higher or equal to 10kbps/timeslot

The reference point is defined tobe able to compare results andestimate capacity.

• QoS target/ref point ����Capacity estimateAt the reference point anaverage of 700 parallell WAPsessions per sector can residebefort the quality limit is reached

Simulation results WAP over EGPRS:Average packet throughput for WAP

0 50 100 150 2000

5

10

15

20

25

30

35

40

45

50

Average offered load [kbps]

Ave

rage

pac

ket t

hrou

ghpu

t per

tim

eslo

t [kb

ps]

(25)

(50)(100)

(200)(300)

(350)(400)

(500)(600)

(700)

(800)

(In parenthesis average number of users per cell)

• At the reference point,where 90% of thepackets have at least10 kbps/timeslot (700parallel users) anaverage throughput of31 kbps/timeslot canbe seen

• With lower loads ahigher averagethroughput is reached

Simulation results WAP over EGPRS:Queuing and transmission delays

0 50 100 150 2000

0.01

0.02

0.03

0.04

0.05

0.06

Average offered load [kbps]

Ave

rage

nor

mal

ised

del

ay p

er p

acke

t [s/

kbit]

total queuing transmission

• The transfer time forWAP increases withhigher load due to moreretransmissions

• The major part of thetotal delay for higherloads is due to queuingtime

Comparing EGPRS vs. GPRS with WAP traffic:Major conclusions

Average Download time for WAP

0,00

0,05

0,10

0,15

0,20

0,25

0,30

0,35

0,40

EGPRS GPRS

Average amount of users and Spectral efficency

0

100

200

300

400

500

600

700

800

EGPRS GPRSAverage number of parallell WAP sessions/sectorSpectral efficiency [kbps/MHz/site]

• EGPRS can handle 4.7times as many usersas GPRS

• EGPRS is 4.5 timesmore spectral efficientthan GPRS

• The time it takes todownload a WAP packetis very small(0.18s for EGPRS and0.36 s for GPRS)

The WAP end user will not perceive major differences in qualityThe system performance increase for the operators are significant

Comparing WAP vs. WWW over EGPRS:Model differences

Packetinterarrival time

Packetsize

Sessioninterarrival time

Sessionlength

WAP WWWMax packet size [Byte] 1600 N/AMean packet size [Byte] 471 8800Standard deviation [Byte] 355 4400Average interarrival time [seconds] 13.8 0.2/3.6/28.4Average session time [seconds] 69 100

Model constants

• WWW have ~19 timeslarger packetsthan WAP

• WWW is a state basedmodel with 3 trafficintensities compared toWAP that isn’t statebase

• WWW models have inaverage 10 packets persession compared to5 for WAP

Comparing WAP vs. WWW over EGPRS:Capacity for two types of Internet browsing

0 100 200 300 400 500 600 700 8000

5

10

15

20

25

30

35

Average number of users per sector

10%

per

cent

ile o

f pac

ket t

hrou

ghpu

t per

tim

eslo

t [kb

ps]

WAPWWW

• In the defined referencepoint an EGPRS carriercan handle ~18 times asmany WAP users asWWW users in thesystem

• At the reference point,where 90% of thepackets have athroughput of at least 10kbps/timeslot, thesystem can host 700WAP users per sectorcompared to 40 WWWusers

Comparing WAP vs. WWW over EGPRS:Spectral efficiency for WAP and WWW

0 50 100 150 200 250 300 3500

5

10

15

20

25

30

35

System load [kbps/MHz/site]

10th

per

cent

ile o

f pac

ket t

hrou

ghpu

t per

tim

eslo

t [kb

ps]

WAPWWW

• The spectral efficiency ishigher for WWW than forWAP

• Note that even though WAPhave 700 user compared to40 WWW users the WWWservice uses the availablebandwidth more efficiently

Reason: WAP is moreeffected by final ACK and thesmall WAP packets leave upto 20 percent empty space inthe RLC blocks

Comparing WAP vs. WWW over EGPRS:Average download time

Average download time [s]

0,00

0,50

1,00

1,50

2,00

2,50

3,00

3,50

4,00

WAP WWW

• The average time todownload a WWWpacket is 3.5 secondscompared to 0.2seconds for WAP

WAP is a service that the have very short waiting periods for the usercompared to WWW

Summary

• WAP over EGPRS and GPRS have similar download performance• WAP over EGPRS can handle 700 WAP users• WAP is a service with good performance over both carriers• WAP can host many more users than WWW (18 times)• WWW is more spectral efficient than WAP• Huge difference in download times for WAP and WWW (19 times)