An Ethernet performance model with N/sub u/ users simultaneously accessing N/sub s/ servers

Abstract

A previously published LAN performance model had two user response time components that involved transmission on the LAN: simultaneous input requests for N/sub u/ users for service from one of N/sub s/ servers transmitted over a broadband Ethernet; and transmission of the requested information over the Ethernet to the N/sub u/ user RAMs. The purpose of the model was to predict user response time under the condition of maximum stress on the LAN (i.e., when all users simultaneously request the same program). In the first implementation of the model, the Ethernet response time components were modeled by classical formulas. Although the model proved to be a good predictor of response time at light and moderate loads, it underestimated response time at high loads. The contribution of the revised model presented to network performance evaluation is that the author has developed a novel way of providing a level of detail and accuracy that other analytic models of Ethernet performance do not provide. A brief analysis of the potential effect on performance in this application environment of the emerging fast Ethernet is provided.<>