QoS of Internet Access with GPRS

Transcription

1 Dept. of Prof. Dr. P. Tran-Gia QoS of Internet Access with GPRS Dirk Staehle 1, Kenji Leibnitz 1, and Konstantin Tsipotis 2 1 [staehle,leibnitz]@informatik.uni-wuerzburg.de 2 Libertel-Vodafone k.tsipotis@libertel.nl

2 Overview Introduction Simulation model traffic model application specific protocols core network model GPRS model Numerical results performance of GPRS comparison to Internet access by modem Conclusion and outlook

3 GPRS GPRS extension of GSM to provide mobile Internet access Higher bandwidth achieved by Usage of multiple channels (PDCHs) Coding schemes with higher data rate Better usage of bandwidth by packet switching GPRS Network Internet Intranet Server Mobile with GPRS

4 Simulation Concept GPRS user GPRS simulation 1 2 m BTS RTT approximation IP network WWW Server FTP WAP WWW FTP-active FTP-idle S A FTP I FTP t A t WWW-active WWW-idle A WWW I WWW t WAP-active A WAP WAP-idle I WAP t traffic model

5 Major Applications Measurements of the student modem access at the Other TCP Other TCP UDP- Games Other TCP HTTP Mail HTTP Mail HTTP FTP Mail 9.6kbps 33.6kbps 64kbps Most important applications: WWW and FTP no major application for GPRS New applications arise (e.g. games, WAP)

6 Traffic Model Applications WWW Literature study of measurements in wired networks H. Choi and J. Limb, Georgia Tech., 1999 A. Reyes-Lecuona, et al, Universidad de Malaga, Spain, 1999 N. Vicari, Universität Würzburg, 1997 P. Barford and M. Crovella, Boston University, 1998 Own measurements Size of s Development of a traffic generator for GPRS users

7 Arrival Process for WWW and InterSessionTime p WWW p WWW WWW Session WWW Session 0.5p Send 0.5p Send 0.5p Receive 0.5p Receive

8 Web Session Model SessionInterArrivalTime WWW Session WWW Session ViewingTime Web Page Web Page Web Page 1 X WWW TCP 4 TCP 3 TCP 2 TCP 1 Main Object Inline Object Inline Object Inline Object Inline Object Inline Object Inline Object Main Object Inline Object Inline Object version of HTTP and browser

9 Parameter Variable Distribution Mean Standard deviation Web request viewing time V WWW Weibull(a,b) truncated at a maximum of 15 min s 92.6 s Number of inline objects N WWW Gamma(0.24, 23.42) Size of main object M WWW Lognormal(1.31, 1.41) 10kB 25kB Size of inline object O WWW Lognormal(-0.75, 2.36) 7.7kB 126kB Size of GetRequest R WWW Lognormal(5.84, 0.29) 360B 106B Number of web requests per WWW Session Lognormal(1.8, 1.68) X WWW

10 Model CDF FUNET Bolotin Measurements in Würzburg size [kbyte]

11 GPRS Simulation Web or Mail Server BTS MS HTTP TCP/IP wired network RLC MAC HTTP TCP/IP RLC MAC air interface

12 Simulation Parameters und Results Simulated scenario: Base station with 3 frequencies Voice calls according to a blocking probability of 1% Coding scheme 2 with RLC block error rate: 7.5% Only on-demand, no dedicated PDCH Mobiles with 4 neighbored slots Results: QoS parameter depending on session arrival rate Mean and median of the time to load a web page Bandwidth of a TCP connection Bandwidth of a TBF Comparison to modem Internet access

13 Data Flow on the Air Interface 120 Transported Traffic [kbps] downlink uplink SessionArrivalRate[1/h]

14 User Activity 100 Active users WWW s downlink TBF receive send SessionArrivalRate[1/h]

15 WWW Downloading Time Downloading time [s] mean median Downloading time [s] mean median SessionArrivalRate[1/h] SessionArrivalRate[1/h]

16 Bandwidth of TCP Connections TCP bandwidth [kbps] WWW receive send SessionArrivalRate[1/h]

17 TBF Bandwidth 50 uplink TBF bandwidth [kbps] per TBF downlink while TBF SessionArrivalRate[1/h]

18 Simulation of Modem Internet Access Modem user wired line access (fixed bandwidth) Access router RTT approximation (as in GPRS simulation) IP network WWW Server FTP WAP WWW FTP-active FTP-idle S A FTP I FTP t A t WWW-active WWW-idle A WWW I WWW t WAP-active A WAP WAP-idle I WAP t traffic model (like GPRS user)

19 GPRS vs. Modem Internet Access Slowdown Factor SessionArrivalRate[1/h] 10 kbps Modem 32 kbps Modem 64 kbps Modem GPRS is better Ratio of WWW downloading times GPRS is worse SessionArrivalRate[1/h] 10 kbps Modem 32 kbps Modem 64 kbps Modem TCP Bandwidth GPRS Slowdown Factor= TCP Bandwidth Modem

20 Conclusion and Outlook Conclusion Stochastic model for GPRS users based on wired network measurements Interaction of TCP and RLC works well RLC retransmissions do not cause TCP timeouts 4 slot mobiles achieve performance of 30kbps-modems for lightly to medium loaded cells Outlook Investigation of Burst errors and Handover on TCP Extension of traffic model WAP Video/audio-streaming Simulation of the UMTS air interface