A goal of modern broadband networks is their ability to provide stringent QoS guarantees to different classes of users. This feature is often related to events with a small probability of occurring, but with severe consequences when they occur. In this paper we focus on the overflow probability estimation and analyze the performance of Bridge Monte-Carlo (BMC), an alternative to Importance Sampling (IS), for the Monte-Carlo estimation of rare events with Gaussian processes. After a short description of BMC estimator, we prove that the proposed approach has clear advantages over the widespread single-twist IS in terms of variance reduction. Finally, to better highlight the theoretical results, we present some simulation outcomes for a single server queue fed by fraction Brownian motion, the canonical model in the framework of long range dependent traffic.
Titolo: | Rare events of Gaussian processes: a performance comparison between Bridge Monte-Carlo and Importance Sampling |
Autori interni: | |
Anno del prodotto: | 2007 |
Rivista: | |
Abstract: | A goal of modern broadband networks is their ability to provide stringent QoS guarantees to different classes of users. This feature is often related to events with a small probability of occurring, but with severe consequences when they occur. In this paper we focus on the overflow probability estimation and analyze the performance of Bridge Monte-Carlo (BMC), an alternative to Importance Sampling (IS), for the Monte-Carlo estimation of rare events with Gaussian processes. After a short description of BMC estimator, we prove that the proposed approach has clear advantages over the widespread single-twist IS in terms of variance reduction. Finally, to better highlight the theoretical results, we present some simulation outcomes for a single server queue fed by fraction Brownian motion, the canonical model in the framework of long range dependent traffic. |
Handle: | http://hdl.handle.net/11568/109296 |
ISBN: | 9783540748328 |
Appare nelle tipologie: | 4.1 Contributo in Atti di convegno |