Large deviations for the empirical mean of an M/M/1 queue

Large deviations for the empirical mean of an M/M/1 queue

0.00 Avg rating0 Votes
Article ID: iaor20132903
Volume: 73
Issue: 4
Start Page Number: 425
End Page Number: 446
Publication Date: Apr 2013
Journal: Queueing Systems
Authors: , ,
Keywords: optimization
Abstract:

Let ( Q ( k ) : k 0 ) equ1 be an M / M / 1 equ2 queue with traffic intensity ρ ( 0,1 ) equ3 . Consider the quantity S n ( p ) = 1 n j = 1 n Q j p equ4 for any p > 0 equ5 . The ergodic theorem yields that S n ( p ) μ ( p ) : = E [ Q ( ) p ] equ6 , where Q ( ) equ7 is geometrically distributed with mean ρ / ( 1 ρ ) equ8 . It is known that one can explicitly characterize I ( ϵ ) > 0 equ9 such that lim n 1 n log P ( S n ( p ) < μ p ε ) = I ε , ε > 0 equ10 . In this paper, we show that the approximation of the right tail asymptotics requires a different logarithm scaling, giving lim n 1 n 1 / ( 1 + p ) log P ( S n ( p ) > μ ( p ) + ε ) = C ( p ) ε 1 / ( 1 + p ) , equ11 where C ( p ) > 0 equ12 is obtained as the solution of a variational problem. We discuss why this phenomenon–Weibullian right tail asymptotics rather than exponential asymptotics–can be expected to occur in more general queueing systems.

Reviews

Required fields are marked *. Your email address will not be published.