Variance Reduction & Efficient Rare Event Simulation

WA15.1 Permutated Estimators for Regenerative Simulations

• Marvin Nakayama; NJIT, Simulation & Modeling Lab., Dept. of CIS, Newark, NJ 07102-1982,; marvin@cis.njit.edu
• James Calvin; NJIT, Simulation & Modeling Lab., Dept. of CIS, Newark, NJ 07102-1982,; calvin@cis.njit.edu

We construct estimators by first running a regenerative simulation using one sequence of regeneration points, permuting the cycles based on another sequence of regeneration points, computing an estimator for the new path and averaging overall possible permutations. We discuss small- and large-sample properties of the permuted estimators.

WA15.2 Adaptive Rare Event Simulation of Telecommunication Networks

• Poul E. Heegaard; Norwegian Univ. of Science & Technology, Dept. of Telematics, Trondheim, N-7034 , Norway; poul.heegaard@item.ntnu.no

To enable the use of importance sampling during simulation of large telecommunication networks, a new adaptive parameter biasing strategy is proposed. Its applicability is demonstrated by simultanous simulation of both teletraffic and dependability aspects of a network with service interaction between different traffic classes, preemptive priorities and rerouting on overload/failures.

WA15.3 Simulation of Fluid Queues with Many Sources

• Ad Ridder; Vrije Univ. Amsterdam, Dept. of Econometrics, de Boelelaan 1105, Amsterdam, 1081 HV , The Netherlands; aridder@econ.vu.nl
• Michel Mandjes; KPN Research, Plning Perform. & Reliability, POB 421, Leidschendam, 2660 AK , The Netherlands; m.r.h.mandjes@research.kpn.com

We consider a finite-buffer fluid queue fed by multiple traffic sources and emptied at constant rate. We derive an asymptotic expression of the most likely path to buffer overflow and use this for applying importance sampling in simulations of the queue to estimate loss probabilities.

WA15.4 Multilevel Splitting for Estimating Rare Event Probabilities

• Paul Glasserman; Columbia Bus. School, 415 Uris Hall, 3022 Broadway, New York, NY 10027-6902,; pglasser@research.gsb.columbia.edu
• Philip Heidelberger; IBM, TJ Watson Research Ctr., PO Box 704, Yorktown Heights, NY 10598 ,; berger@watson.ibm.com research.ibm.com/people/b/berger/
• Perwez Shahabuddin; Columbia Univ., 331 SW Mudd Building, IEOR Dept., New York, NY 10027 ,;
• Tim Zajic; Columbia Univ., IEOR Dept., NY, NY 10027 ,;

We analyze a multilevel splitting technique for estimating rare event probabilities. For a particular class of models, this method is asymptotically optimal, provided the splitting factor is properly chosen. However, for more general models, the method is not efficient unless the splitting is consistent with the model's large deviations behavior.