Jean-Sébastien Tancrez,
Philippe Chevalier and Pierre Semal. Modelling Queueing
Networks with Blocking using Probability Masses Fitting (abstract only)
information
Abstract. In this paper, we are interested in the modelling
of queueing networks with split and merge operations,
and without loops. Such a network maybe used to model assembly/disassembly
manufacturing systems, for example. The buffers are finite, and can thus lead
to blocking in the network. No assumption is taken on service time
distributions.
Such queueing networks, with general service time
distributions, cannot be modeled exactly. In order to use analytical models,
the distributions have to be transformed to “tractable” distributions, i.e.
phase-type distributions most of the time. Our originality mainly lies in this
stage of the modelling process. We propose
“probability masses fitting”, to build tractable
discrete phase-type distributions. Then, the queueing
network can be modeled by a discrete Markov chain, from which the performance
measures can be deduced.
Probability masses fitting (PMF) is quite intuitive: the probability masses on
regular intervals are computed and aggregated on a single value in the
corresponding interval, leading to a discrete distribution. PMF has some
interesting characteristics: it conserves the shape of the distribution (but
not the moments, as do moments fitting), it can be refined, and it is particularly
natural when the data about the processing time distributions is collected in
the form of histograms.
Probability masses fitting leads to various results. First, it allows to compute lower and upper refinable
bounds on the throughput of the network. Second, the distribution of the cycle
time can be estimated and shows to be realistic. Finally, we show by
computational experiments that good accuracy levels are reached in the
evaluation of various performances (cycle time, work-in-progress, flow time).