The persistent congestion problem of FAST-TCP: analysis and solutions

TL;DR

This paper analyzes the persistent congestion problem in FAST-TCP, models its unfairness issues, and proposes a sender-side algorithm to improve fairness without router modifications.

Contribution

It introduces a model for FAST-TCP's unfairness, and develops a sender-based delay estimation algorithm to enhance fairness while maintaining protocol efficiency.

Findings

The model accurately predicts unfairness among multiple FAST-TCP flows.

The proposed algorithm improves fairness between old and new connections.

Fairness is achieved without changing network infrastructure.

Abstract

FAST-TCP achieves better performance than traditional TCP-Reno schemes, but unfortunately it is inherently unfair to older connections due to wrong estimations of the round-trip propagation delay. This paper presents a model for this anomalous behavior of FAST flows, known as the persistent congestion problem. We first develop an elementary analysis for a scenario with just two flows, and then build up the general case with an arbitrary number of flows. The model correctly quantifies how much unfairness shows up among the different connections, confirming experimental observations made by several previous studies. We built on this model to develop an algorithm to obtain a good estimate of the propagation delay for FAST-TCP that enables to achieve fairness between aged and new connections while preserving the high throughput and low buffer occupancy of the original protocol. Furthermore, our proposal only requires a modification of the sender host, avoiding the need to upgrade the intermediate routers in any way.