Abstract

This paper focuses on the performance of the TCP protocol [Pos81], which is used for reliable, connectionoriented transfer of information. Since TCP will interpret a bit error introduced during a link fade as an indication of congestion, it is important to understand the performance implications of allowing the satellite link to operate at a non-zero bit error rate. Mitigation of fade-induced bit errors carries a cost in the form of increased power requirements or the need for error correction, which requires additional bandwidth and circuitry. Allowing the bit error rate on the link to rise too much will leave the satellite user with poor performance, and use link capacity for retransmission of damaged packets. It is therefore important to understand the maximum error rate at which TCP can operate with acceptable performance. This paper reports on an experimental investigation of this question using the NASA ACTS satellite during user-induced fades. The focus of this study is twofold. First, we examine TCP configurations that are likely to

Citations