Application-perceived throughput plays a major role for the performance of networked applications and thus for user experience. Especially in mobile environments, throughput is limited, which is a challenge for applications and users. On this background, this tutorial paper investigates the process of user-perceived throughput in GPRS and UMTS systems seen over rather small averaging intervals. It discusses the corresponding active measurements mimicking the needs of streaming applications and analyzes results with aid of summary statistics, histograms and autocorrelation coefficients. These results reveal a clear influence of the network, seen from variations and autocorrelation of application-perceived throughput mostly on the one-second time scale. Such an influence has to be considered when choosing the right kind of network for a specific task. To that aim, the investigated metrics will serve as input to a multi-criteria decision process to determine the proper network for a specific service. The results indicate that, applications have to cope with considerably large jitter when trying to use the nominal throughputs. In GPRS, the promised average throughputs are not even reached in downlink direction; instead, a quite large amount of packet loss occurs. With aid of causality arguments for an equivalent bottleneck, bounds for the extra delay of the first packet sent via mobile links can in fact be derived from throughput measurements.

En grundlig redovising av throughput-mätningar och deras innebörd i mobila miljöer; samtidig en demonstration att den ekvivalenta flaskhalsmodellen kan användas för fördröjningsuppskattningar