Impact of data traffic composition on GPRS performance Hamza Dahmouni 1 , David Rossé 1 , Bertrand Morin 1 and Sandrine Vaton 2 1 France Telecom Research and Development, 6 avenue des usines - 90007 Belfort France {hamza.dahmouni,david.rosse,bertrand.morin}@francetelecom.com , 2 ENST-Bretagne, Dept. of Computer Science - CS 83818 - 29238 Brest Cedex 3, France sandrine.vaton@enst-bretagne.fr Abstract: The topic of this paper is the evaluation of the impact of data traffic composition on GPRS performance. A detailed statistical analysis of GPRS traffic per service at different levels has been carried on live measurements. A multi-service performance model is presented in order to derive simple and closed form formulas, associating capacity, traffic characteristics and performance. This performance model allows the analysis of the impact of a mix of the applications (WAP, Web, e-mail, etc…) on the quality of service QoS offered to GPRS users. It also gives the opportunity to evaluate the impact of the introduction of new services on the QoS offered to already existing services. Keywords: traffic model, GPRS, live measurement, processor sharing, performance model 1. INTRODUCTION The main challenge in the engineering of a multi-service mobile network is the integration and support of a wide variety of services such as WAP, MMS, Web, e-mail, etc. These data services require different levels of quality of service (QoS). Moreover the traffic corresponding to each service is highly bursty and can be characterized by a typical ON/OFF process [5]. The total traffic generated by data users is consequently highly dependent on the burstiness of each service. Optimal dimensioning of this multi-service network requires the knowledge of the traffic characteristics of each service. The expected merge of the wireline and the wireless Internet implies that a dramatic change will occur in traffic composition within mobile networks. For example, Web traffic will probably overflow WAP traffic. Indeed, HTML compatible browsers will soon be widely available in mobile handsets. And the most popular wireless application WAP will emerge towards TCP, following WAP 2.0, and will show characteristics similar to the Web service. In order to analyze how the content of the applications mix (e.g. WAP, Web) impacts the QoS of the GPRS networks, and also to evaluate the impact of the introduction of a new service on the performance of existing services, we have developed a multi-service performance model. This model relies on an ON/OFF model for each service. It requires a realistic characterization of the duration of the OFF-period and of the size of the ON-period. That is why a traffic characterization has been carried out on live traces of the Orange GPRS network. The analysis of these live measurements was conducted at different levels (packet level and flow level) for various services supported by GPRS. Even though Internet traffic characterization has been a subject of interest for many years, very few results have been published to date on traffic characterization for wireless networks. Existing works