This paper deals with the application of Generalized Extreme Value Theory (GEVT) to the ATM buffer dimensioning. In particular, a general ATM system simulation model has been defined in order to analyze a large variety of buffer behaviours. The performance of the classical GEVT estimate procedure are evaluated in some critical cases by stressing its limits. In order 10 better cope with these cases, a modified version of the GEVT estimate procedure is proposed. Such a procedure is based on the combined utilization of the knowledge obtained from the classical Montecarlo method and the extrapolating form of the GEVT estimator. The obtained results show that this method allows some limitations of the GEVT approach to be overcome and a very good result accuracy can be reached.
A combined Montecarlo/GEVT extrapolating method for the estimate of buffer length distribution tails / F., Bernabei; R., Ferretti; Listanti, Marco; G., Zingrillo. - STAMPA. - (1992), pp. 1206-1211. (Intervento presentato al convegno International Conference on Communications 1992 tenutosi a Chicago (USA) nel June 1992) [10.1109/ICC.1992.268049].
A combined Montecarlo/GEVT extrapolating method for the estimate of buffer length distribution tails
LISTANTI, Marco;
1992
Abstract
This paper deals with the application of Generalized Extreme Value Theory (GEVT) to the ATM buffer dimensioning. In particular, a general ATM system simulation model has been defined in order to analyze a large variety of buffer behaviours. The performance of the classical GEVT estimate procedure are evaluated in some critical cases by stressing its limits. In order 10 better cope with these cases, a modified version of the GEVT estimate procedure is proposed. Such a procedure is based on the combined utilization of the knowledge obtained from the classical Montecarlo method and the extrapolating form of the GEVT estimator. The obtained results show that this method allows some limitations of the GEVT approach to be overcome and a very good result accuracy can be reached.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
