B. Measurements
The traffic generator developed for the measurements
models a WWW/e-mail traffic mix. A session generator
randomly creates e-mail (70%) or WWW (30%) sessions.
E-mail sessions comprise one e-mail of on average 10 kbyte
from a mail server. A web session consists of five pages,
with an average of 2.5 objects. Each object has an average
size of 3.7 kbyte. Since it was not possible to generate
files on the server dynamically, these settings were reproduced
by five files on the server with sizes of 1, 2, 2.5, 5 and
10 kbyte. The idle time between two sessions as well as the
read time per web page was set to twelve seconds. Thus, the
traffic is comparable to the GPRSIM traffic models regarding
the mean values, although it does not model the same
exact statistic behavior. Some differences remained in contrast
to the GPRSIM. The small delay of less than 40 ms
in external networks is neglected in the simulator. The delay
between two web objects was between 3 and 4 seconds,
which is zero in the GPRSIM. The GPRSIM also does not
model POP3 authentication.
C. Comparison to GPRSIM Results
The traffic performance evaluated by the GPRSIM is
significantly higher than in the measurements (see Figure
10). The main reason for the difference of the throughput
performance in the real network and the GPRSIM results
for this traffic scenario is the high round-trip time at
the radio interface of the real network at that time (see Section
III), which slows down the throughput dramatically especially
for small files that are regarded here. This high
signaling delay in the real network will be enhanced in the
next versions of the protocol software in mobile terminals
and base stations in the future [10]. The GPRSIM can be
regarded as representative for optimized GPRS equipment.