In this paper we present a new TCP startup algorithm, called Paced Start (PaSt). It builds on recent results in the area of available bandwidth estimation to automatically determine a good value for ssthresh. Paced Start is based on the simple observation that the TCP startup packet sequence is in fact a sequence of packet trains. By monitoring the difference between the spacing of the outgoing data packets and the spacing of the incoming acknowledgements, Paced Start can quickly estimate the proper congestion window for the path. To examine TCP performance over cellular data networks using various techniques to detect/infer the existence of such performance-enhancing mechanisms. Special attention is given to TCP split approaches. The characterize packet-level loss on the cellular link, and discuss its impact on the end-to-end throughput of TCP connections and also performed experiments using UDP probes to examine the artifacts of wireless channels. Here found that the wireless channels are asymmetric in several aspects such as call setup delay, loss rate, packet concatenation, and packet re-ordering. The analysis results based on this model convince us that the spurious timeout caused by abrupt bandwidth decrease is one of the most important factors which degrade TCP performance. In addition, the loss of synchronization (even conflict) between the congestion window adjustment and the bandwidth oscillation results in bandwidth waste and buffer overflow which leads bursty packet drops.
展开▼