A key requirement of modern telecommunications networks is the capability of providing a variety of Quality of Service (QoS) guarantees to different classes of users. In this scenario scheduling systems play a major role, thanks to their ability of offering different levels of service, while preventing some "greedy users" from degrading the QoS of other classes. A primary QoS parameter is the Cell Loss Probability (CLP), whose typical values are very small and therefore difficult to estimate through standard Monte Carlo (MC) simulation, since long run times are required to achieve accurate results. Under proper conditions. Importance Sampling (IS) techniques can speed up simulations involving rare events. In this paper we propose an application of IS to the simulation of a Generalized Processor Sharing (GPS) scheduler, fed by Markov Arrival Processes (MAP). The algorithms we propose are based on large deviations principles, which provide asymptotic results on the decay rate of per-session queue length tail distributions in an idealized GPS discipline. In particular, we first present a scheme for the simulation of a two-buffer system, then we extend the technique to the general case of a multi-buffer scheduler. We subsequently employ our algorithms to simulate a packetized realistic system, namely the Packet-by-packet Generalized Processor Sharing.
An Importance Sampling Algorithm for the Simulation of a GPS scheduler
GIORDANO, STEFANO;PAGANO, MICHELE;
2002-01-01
Abstract
A key requirement of modern telecommunications networks is the capability of providing a variety of Quality of Service (QoS) guarantees to different classes of users. In this scenario scheduling systems play a major role, thanks to their ability of offering different levels of service, while preventing some "greedy users" from degrading the QoS of other classes. A primary QoS parameter is the Cell Loss Probability (CLP), whose typical values are very small and therefore difficult to estimate through standard Monte Carlo (MC) simulation, since long run times are required to achieve accurate results. Under proper conditions. Importance Sampling (IS) techniques can speed up simulations involving rare events. In this paper we propose an application of IS to the simulation of a Generalized Processor Sharing (GPS) scheduler, fed by Markov Arrival Processes (MAP). The algorithms we propose are based on large deviations principles, which provide asymptotic results on the decay rate of per-session queue length tail distributions in an idealized GPS discipline. In particular, we first present a scheme for the simulation of a two-buffer system, then we extend the technique to the general case of a multi-buffer scheduler. We subsequently employ our algorithms to simulate a packetized realistic system, namely the Packet-by-packet Generalized Processor Sharing.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
