Papers and Reports on Half-Duplex Ethernet Performance
The papers and reports listed below are maintained here for historical purposes only. These papers and reports only apply to half-duplex Ethernet systems and at as such are only of interest to historians of technology.
The half-duplex Ethernet system was the subject of a number of papers and reports on its performance. The performance of the half-duplex system could be complex to model due to the shared medium and random access of computers connected to the system. All modern Ethernet systems use full-duplex, which eliminates the use of a shared medium and also eliminates the use of the CSMA/CD access control system to the shared medium. Therefore, all full-duplex Ethernet systems can operate to the full bandwidth of the Ethernet link, since there are only two devices on each full-duplex link, one at each end.
Note: The following materials are about half-duplex Ethernet only, and have nothing to do with modern full-duplex Ethernet systems.
When it comes to real-world network throughput, there are as many opinions about Ethernet performance as there are Ethernet installations. Every network site is unique, and the performance of the network is a function of the number of stations contending for access to the Ethernet, the type of hardware in use (station interfaces, switches), the layout of the cables (are they the right type, right length, and do they meet the config guidelines?), the quality of the cable installation, the mix of applications, and so on.
The following items help demonstrate that an Ethernet LAN can transfer data at quite close to the nominal rated speed of the technology.
A New Binary Logarithmic Arbitration Method for Ethernet. This paper by Dr. Mart Molle provides a very complete description of Ethernet performance and documents tests in which the Ethernet channel was driven to 100% load. The paper also describes the Ethernet channel capture effect and includes a solution for channel capture based on modifying the Ethernet MAC. Note that some pages are missing from this file. The missing pages are composed of a set of full page figures that accompany the paper, which can be retrieved by clicking here:
Full page figures for Molle paper
Measured Capacity of an Ethernet: Myths and Reality, by Boggs, Mogul, and Kent. From the abstract:"Based on measurements of an actual implementation, we show that for a wide class of applications, Ethernet is capable of carrying its nominal bandwidth of useful traffic, and allocates the bandwidth fairly. We discuss how implementations can achieve this performance, describe some problems that have arisen in existing implementations, and suggest ways to avoid future problems."
Ethernet Channel Capture Ethernet channel capture is a phenomenon in which the half-duplex Ethernet media access control (MAC) system can become biased for a short term toward one station on a heavily loaded network. Under certain circumstances, this allows a station to more frequently win the contention for the channel, or ``capture'' the channel, while that station has something to send. Usenet posting on Capture Effect. A short description of how capture effect works can be found in this Usenet posting from Rich Seifert.
The Packet Starvation Effect in CSMA/CD LANs Part 1 The Packet Starvation Effect in CSMA/CD LANs Part 2
A paper on the "Packet Starvation Effect," and its effects on network latency for applications like packet video. "Packet starvation" is another way of describing the effects of Ethernet channel capture.
The Effect of Ethernet Behavior on Networks using High-Performance Workstations and Servers.
(Adobe PDF version - approx 96 KBytes).
A technical report from Rich Seifert, one of the original designers of Ethernet, describing how Ethernets function in the presence of high offered load from powerful workstations. Includes a section on the Ethernet capture effect.
Issues in LAN Switching and Migration from a Shared LAN Environment.
(Adobe PDF version - approx 519 KBytes).
In this technical report Rich Seifert describes how switched Ethernets function and what issues to consider when planning to implement switched Ethernets at your site.
Usenet postings about Ethernet collisions. Two perennial questions about Ethernet are: What is a collision? and, What rate of collisions are acceptable? These questions are discussed in this set of Usenet postings.
What is SQE Test and When to Use It. The question: "What is the SQE Test signal and when should it be enabled on outboard transceivers?" is discussed in this document