Asymmetric uplink and downlink rates are common in many broadband access networks, particularly DSL, DOCSIS, and cellular. With the advent of peer-to-peer protocols and pervasive devices supporting multimedia capabilities, bi-directional TCP traffic has also become commonplace [1]. Under conditions where TCP connections are sending data in both directions simultaneously, performance, as we will show, can be severely degraded.

This thesis seeks to determine how buffers, combined with elementary queueing strategies like FIFO, influence bi-directional TCP performance across asymmetric network links. It also seeks to determine if there is a way to size buffers such that the throughput for a small number of bi-directional connections is optimized.

To answer this question, fundamental system components and algorithms will be reviewed. Following this, research pertaining to observations, optimization techniques and design spaces will be analyzed. Next, a set of real world experiments will be analyzed. Finally, a buffer sizing model will be proposed that optimizes throughput for the bi-directional, long lived, TCP traffic under a FIFO queueing strategy.

Practically speaking, this thesis seeks to determine if current consumer grade customer premise equipment is designed to provide adequate performance for bi-directional Internet traffic across typical asymmetric access technologies like DSL and DOCSIS.