摘要
This work explores the behavior of both TCP-Reno and TCP-Sack under a simple scenario, where a single TCP source transmits the packets continuously over a single bottleneck node characterized by its queue size, bandwidth and propagation delay. The analysis allows to derive the performance of TCP, the utilization tends to 75% of the bottleneck throughput when the bandwidth×propagation delay pipe becomes very large, while it tends to 100% when the queuing delays are predominant because the queue is never empty. In the transient analysis we show how the initial phase of the session can degrade the performances. These results are proved through simulation.
This work explores the behavior of both TCP-Reno and TCP-Sack under a simple scenario, where a single TCP source transmits the packets continuously over a single bottleneck node characterized by its queue size, bandwidth and propagation delay. The analysis allows to derive the performance of TCP, the utilization tends to 75% of the bottleneck throughput when the bandwidth×propagation delay pipe becomes very large, while it tends to 100% when the queuing delays are predominant because the queue is never empty. In the transient analysis we show how the initial phase of the session can degrade the performances. These results are proved through simulation.
基金
the National Natural Science Fundation of China!49402004
