Evaluation of Clustering Techniques to Estimate the Effective Bandwidth of a Markovian Fluid from Traffic Traces

Integrated services digital networks, designed to transport data in real time, are modeled by a multiplexer system, where several data fluids share a single output. At the time of admission of a new connection, in order to maintain the quality of service (QoS), it is important to know the amount of...

Full description

Saved in:
Bibliographic Details
Published in:Revista IEEE América Latina 2023-05, Vol.21 (5), p.636-642
Main Authors: Fernandez, Carina, Bavio, Jose, Marron, Beatriz
Format: Article
Language:eng ; por ; spa
Subjects:
Bandwidth
Bandwidths
Clustering
Data models
Effective bandwidth
Entropy (Information theory)
Fluids
Markovian fluid
Mathematical models
Maximum likelihood estimation
Media
Multiplexing
Parameters
Quality of service
Random variables
Resource allocation
Supervised and unsupervised estimation methods
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Integrated services digital networks, designed to transport data in real time, are modeled by a multiplexer system, where several data fluids share a single output. At the time of admission of a new connection, in order to maintain the quality of service (QoS), it is important to know the amount of available resources required by the connections sharing the channel. It is therefore important to have models for the sources and techniques that provide accurate estimates of the resources required for each of them. It is assumed that the network sources are modeled using the Generalized Markovian Fluid Model (GMFM) because of its versatility in describing traffic fluctuations. This is a Markovian fluid where the transfer rate is a random variable whose range and probability distribution are determined by the state of the modulating chain. To measure resource allocation, the concept of Effective Bandwidth (EB) is used, since it allows expressing this magnitude as a function of the model parameters, which will be estimated from traffic traces. As the size of the data to describe the behavior of a source is clearly enormous, they are studied using clustering techniques. In this work, different methods, supervised and unsupervised, are presented to estimate the parameters involved in the calculation of the EB. Finally, the performance of the estimators is analyzed by calculating partition comparison indices, corresponding to the dispatch intervals, which are based on the confusion matrix and the notion of mutual information entropy between partitions, on simulated data.
ISSN:1548-0992
1548-0992
DOI:10.1109/TLA.2023.10130835