2002Unpublished venueRequires access

Performance analysis of TCP in the presence of random losses/errors

Aldar C.-F. Chan, Danny H. K. Tsang, S. Gupta

Open publisher page 4 citations

Abstract

The performance of TCP in the presence of random losses/errors is examined. TCP is originally designed to operate over networks which are error/loss free. However, this assumption is violated in many networks (wireless, ATM) over which TCP operates nowadays. The effects of TCP segment losses/errors on TCP sessions are investigated through many simulation experiments. The observations from the simulation traces provide a basis for the constructions of two analytical models (a simple analysis and an exact analysis) for end-to-end TCP throughput evaluation. For WAN connections (with large bandwidth-delay product), both models give close approximates. However, for LAN connections (with small bandwidth-delay product), the simple analysis estimates deviate from simulation results whereas the exact analysis still gives close estimates. The analytical framework developed has applications in studying the effects of TCP parameters, channel characteristics, intermittent connectivity due to handoff on TCP throughput, through which we can determine how TCP needs to be modified in the long term for efficient operation over emerging networks which are lossy/noisy.

About this research paper

What this paper is about

The performance of TCP in the presence of random losses/errors is examined. TCP is originally designed to operate over networks which are error/loss free. However, this assumption is violated in many networks (wireless, ATM) over which TCP operates nowadays. The effects of TCP segment losses/errors on TCP sessions are investigated through many simulation experiments. The observations from the simulation traces provide a basis for the constructions of two analytical models (a simple analysis and an exact analysis) for end-to-end TCP throughput evaluation. For WAN connections (with large bandwidth-delay product), both models give close approximates. However, for LAN connections (with small bandwidth-delay product), the simple analysis estimates deviate from simulation results whereas the exact analysis still gives close estimates. The analytical framework developed has applications in studying the effects of TCP parameters, channel characteristics, intermittent connectivity due to handoff on TCP throughput, through which we can determine how TCP needs to be modified in the long term for efficient operation over emerging networks which are lossy/noisy.

Why it matters

OpenAlex reports 4 citations for this work. Citation counts describe recorded attention and do not establish research quality.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

The performance of TCP in the presence of random losses/errors is examined. TCP is originally designed to operate over networks which are error/loss free. However, this assumption is violated in many networks (wireless, ATM) over which TCP operates nowadays. The effects of TCP segment losses/errors on TCP sessions are investigated through many simulation experiments. The observations from the simulation traces provide a basis for the constructions of two analytical models (a simple analysis and an exact analysis) for end-to-end TCP throughput evaluation. For WAN connections (with large bandwidth-delay product), both models give close approximates. However, for LAN connections (with small bandwidth-delay product), the simple analysis estimates deviate from simulation results whereas the exact analysis still gives close estimates. The analytical framework developed has applications in studying the effects of TCP parameters, channel characteristics, intermittent connectivity due to handoff on TCP throughput, through which we can determine how TCP needs to be modified in the long term for efficient operation over emerging networks which are lossy/noisy.

Key concepts: TCP global synchronization, TCP tuning, TCP acceleration, HSTCP, Zeta-TCP, Computer science, TCP Friendly Rate Control, H-TCP

Related papers

Back to paper searchBrowse research topicsOriginal source
Performance analysis of TCP in the presence of random losses/errors — Research Paper | ScholarLens