Nrl
Publications at NRL

Search by Title

Search:

Search by Author

Search:


Conference Paper

Title

Buffer Estimate Filtering Using Dispersion Deltas

Abstract

In this paper, we present a sender-side-only filtering method to improve the accuracy of buffer estimation. Buffer estimates are of interest in many applications, e.g. enhancing the performance and behavior of TCP, as well as improving the user experience of UDP applications. A conventional approach to estimate the buffer size of the bottleneck link is to estimate the packet round trip time (RTT) values that include the maximum and minimum queuing delays. By calculating the difference of these two values, one can then estimate the buffer size. However, such approach is shown to be problematic in the presence of lossy links, bursty cross traffic, and multiple bottleneck links. This paper introduces a new technique called Dispersion Delta Filtering (DDF) which takes not only the RTT range but also packet dispersion into consideration. The dispersion data is used to filter out RTT samples affected by bursty cross traffic. By filtering these nonrepresentative RTT samples, the proposed method is able to estimate the buffer more accurately under bursty cross traffic scenarios. We augment TCPW-BBE (TCP Westwood with Buffer and Bandwidth Estimation) to evaluate the accuracy of a buffer estimate with and without the proposed filtering algorithm. Simulation results show that by incorporating the proposed filtering technique, TCPW-BBE obtains a more accurate buffer estimate regardless of the rates of bursty cross traffic, the direction of bursty cross traffic, as well as the number of coexisting flows. The overall TCPW-BBE throughput is improved while greatly increasing friendliness with TCP-NewReno.

Download
Paper: PDF file of paper

Information & Date

In International Workshop on Protocols for Future, Large-Scale and Diverse Network Transports (PFLDNeT 2009), Tokyo, Japan, May. 2009

Authors

Brandon Pancost
Chine-Chia Chen
M. Y. Sanadidi
Mario Gerla