TCP startup performance in large bandwidth delay networks

Next generation networks with large bandwidth and long delay pose a major challenge to TCP performance, especially during the startup period. In this paper we evaluate the performance of TCP Reno/Newreno, Vegas and Hoe's modification in large bandwidth delay networks. We propose a modified Slow-start mechanism, called Adaptive Start (Astart), to improve the startup performance in such networks. When a connection initially begins or re-starts after a coarse timeout, Astart adaptively and repeatedly resets the Slow-start Threshold (ssthresh) based on an eligible sending rate estimation mechanism proposed in TCP Westwood. By adapting to network conditions during the startup phase, a sender is able to grow the congestion window (cwnd) fast without incurring risk of buffer overflow and multiple losses. Simulation experiments show that Astart can significantly improve the link utilization under various bandwidth, buffer size and round-trip propagation times. The method avoids both under-utilization due to premature Slow-start termination, as well as multiple losses due to initially setting ssthresh too high, or increasing cwnd too fast. Experiments also show that Astart achieves good fairness and friendliness toward TCP NewReno. Lab measurements using a FreeBSD Astart implementation are also reported in this paper, providing further evidence of the gains achievable via Astart.