On using the hyperbolic distribution to statistically model network LAN traffic time-segments

Document Type

Article

Publication Date

12-1-2003

Abstract

Efficient and accurate statistical models of network activity and resource usage, as well as associated performance measures, are useful in the engineering process of optimizing network design. In this paper, we introduce a new statistical model, the hyperbolic distribution, for network resource utilization and performance parameters, that promises very accurate representation in a compact manner. This model has been tested in modeling several performance parameters in data traces collected from actual NJIT campus LANs as well as simulated client server LANs with IP traffic. By fitting the model parameters to the experimental data, our simulation results show that this analytic model performs significantly better than the models currently in common use, such as the Weibull, Pareto, normal and lognormal distributions. It was shown that satisfactory modeling results can be obtained only when the NJIT Campus LAN traffic is segregated into at least three contiguous segments in time, corresponding roughly to the Day, Evening and Night time periods.

Identifier

4944226565 (Scopus)

ISBN

[9608052866, 9789608052864]

Publication Title

Recent Advances in Communications and Computer Science

First Page

358

Last Page

363

This document is currently not available here.

Share

COinS