In recent years, IP networks have become increasingly complex and the range
of network applications has widened. As a result, the need for experimenta
l evaluation of application performance has increased. We introduce a measu
rement-based performance evaluation method that combines a function to gene
rate various patterns of application traffic with one to measure end-to-end
network performance at the application level. For our method, we also prop
ose traffic models of various applications that can represent the character
istics of these applications, for example, the burstiness of traffic. This
method has already been implemented as a benchmarking tool on UNIX operatin
g systems.