We consider a model of a multiclass make-to-stock manufacturing system. Ext
ernal demand for each product class is met from the available finished good
s inventory; unsatisfied demand is backlogged. The objective is to devise a
production policy that minimizes inventory costs subject to guaranteeing s
tockout probabilities to stay bounded above by given constants epsilon (j),
for each product class j (service level guarantees). Such a policy determi
nes whether the facility should be producing (idling decisions), and if it
should, which product class (sequencing decisions). Approximating the origi
nal system, we analyze a corresponding fluid model to make sequencing decis
ions and employ large deviations techniques to make idling ones. We conside
r both linear and quadratic inventory cost structures to obtain a priority-
based and a generalized longest queue first-bused production policy, respec
tively. An important feature of our model is that it accommodates autocorre
lated demand and service processes, both critical features of modem failure
-prone manufacturing systems.