It is well-known that TCP performs poorly in the presence of packet losses happening for reasons other than Internet congestion. One increasingly important source of such errors are wireless links. In this paper, networks are studied where the congestion—and thus the bandwidth available to a host—varies over time, and where in addition packets are lost at random. First, we propose a network comprising both dynamically varying congestion and random errors. Second, our model is extended with network calculus concepts in order to study bursty changes of congestion. Finally, we present the transfer protocol TCP “Wichita” (TCPW) which achieves a provable worst-case performance in this environment.

Top