A stream scheduling scheme based on local regularity of internet traffic / Esquema de escalonamento baseado na regularidade local de fluxos de dados internet

AUTOR(ES)
DATA DE PUBLICAÇÃO

2006

RESUMO

Today network traffic is composed of many services with different statistical characteristics and quality of service requirements. This integration needs efficient traffic congestion control and management schemes. Dynamic and preventive schemes usually anticipate traffic conditions by means of a prediction process. Nevertheless, at fine-grained time scales, traffic exhibits strong irregularities and more complex scaling law that make this prediction process a non-trivial task. In this work we model network traffic flows as multifractal processes and introduce the pointwise Hölder exponent as an indicator of the local regularity degree. Also we propose a new traffic flow scheduling scheme based on the Generalized Processor Sharing (GPS) discipline that incorporate the pointwise Hölder exponent to locally characterize each data flow. For this end we explicitly present both dynamic pointwise Hölder exponent estimation and prediction mechanisms. The pointwise Hölder estimation is carried out dynamically based on the decay of the wavelet coefficients in the selected time windows. The proposed predictor is adaptive and implemented with both Kalman and Normalized Least Mean Squares (NLMS) filters. Experimental evaluations have validated the proposed scheduling scheme, resulting in low data loss rate and a better sharing of the network resources in comparison with the usual GPS scheme

ASSUNTO(S)

filtragem de holder exponent multifractals kalman filter internet (redes de computação) telecomunicações - trafego network traffic kalman wavelets (matematica) scheduling

ACESSO AO ARTIGO

Documentos Relacionados