Long-Haul TCP vs. Cascaded TCP

Feng, W. (2006) Long-Haul TCP vs. Cascaded TCP. Technical Report TR-06-04, Computer Science, Virginia Tech.

Abstract

In this work, we investigate the bandwidth and transfer time of long-haul TCP versus cascaded TCP [5]. First, we discuss the models for TCP throughput. For TCP flows in support of bulk data transfer (i.e., long-lived TCP flows), the TCP throughput models have been derived [2, 3]. These models rely on the congestion-avoidance algorithm of TCP. Though these models cannot be applied with short-lived TCP connections, our interest relative to logistical networking is in longer-lived TCP flows anyway, specifically TCP flows that spend significantly more time in the steady-state congestion-avoidance phase rather than the transient slow-start phase. However, in the case where short-lived TCP connections must be modeled, several TCP latency models have been proposed [1, 4] and based on these latency models, the throughput and transfer time of short-lived TCP connections are obtainable. Using the above models, the transfer times for a data file of size S packets can be computed for both long-haul TCP and cascaded TCP. The performance of both systems is compared via their transfer times. One system is said to be preferred if its tranfer time is lower than that of the other. Based on these performance comparisons, we develop a decision model that decides whether to use the cascaded TCP or long-haul TCP.