We discuss in this article the different issues to be considered when model
ing the TCP protocol in a real environment. The discussion is based on meas
urements we made over the Internet. We show that the Internet is so heterog
eneous that a simplistic assumption about TCP congestion control or the net
work may lead to erroneous results. We outline some of our results in this
field, and we present a novel approach for a correct validation of a model
for TCP.