On Modification of Regenerative Splitting for Embedded Markov chain
Alexandra V. Borodina (Karelian Research Center RAS and Petrozavodsk University, Russia)
We consider the problem of speed-up simulation of stationary probability
,
where is stationary and regenerative queue-size process. For
large level L this probability ought to be small (about 10-20
and lower).
For small probability simulation we use some new technique which includs regeneration method advantages and idea of splitting the trajec- tories of the process.
In the case of nonmarkovian process we have some uncomplited servece time at the splitting moment. As a consequence, this fact can give some reason for increase in variation. To avoid this situation in the queue M/G/1 we have applied the embedded Markov chain method.
Results of experiments have confirmed the assumption of variance reduction, iven for subexponential service time distribution.
The work is supported by the Russian Foundation for the Basic Research, Grant 07-07-00088.
