Abstract-This paper presents a performance evaluation of GPRS accomplished by combination of measurement at the end hosts and tracing inside the network. The multi-layer tracing approach allows not only observing, but also understanding the network performance. With end-to-end measurements we assess data rates, latency, and buffering experienced by users in a live GPRS network. Comparing the results to our previous measurements shows a notable improvement in the network and terminals over past two years. Mobility tests while driving in the urban environment quantify the interval, duration and data loss caused by cell reselections. In the test lab, multi-layer tracing of radio, link and transport protocols gives a closer picture of GPRS performance. For instance, TCP interacts inefficiently with resource allocation at the RLC layer and fragmentation at the LLC layer. Finally, we illustrate delay spikes and data losses during a cell reselection by tracing of signaling messages during a cell update and routing area update procedures.

-In this paper we present results of measurements on the performance of GSM HSCSD and GPRS data transmission. We used a measurement tool that we have developed to study the performance of various wireless links as perceived by nomadic applications using the TCP protocol. The results show that in stationary connections the throughput and response time were stable and, in general, close to the theoretical values. However, the throughput and response time varied a lot when connections were used in motion. One of the reasons is that TCP was not capable to adapt itself properly to the variability of QoS of HSCSD and GPRS, and therefore, it did a lot of unnecessary retransmissions causing performance slowdown. The performance of HSCSD was better than the performance of GPRS. Reliability was adequate in stationary connections, but in moving connections there were unwanted disconnections or long pauses in data transfer.