Structured buffer-allocation problems (Q1911468)

The paper shows that some general and quite intricate models of controlled production lines may be represented as specially structured GSMPs (generalized semi-Markov processes). The framework of GSMPs is first reviewed with particular attention being paid to some relevant results. Then, a general blocking model (incorporating limited intermediate finished goods inventory at each stage, as well as limited intermediate raw material inventory) is presented and formulated as a GSMP. Following this, the effect of changing buffer sizes in serial lines with general blocking is analysed. More precisely a reversibility property, emerging from a general representation of event epochs in GSMPs, is established that cuts in half the number of candidate allocations since each allocation has the same throughput as its reverse. Moreover, related results are derived for a system with constant work-in-progress. For the case of synchronized service, i.e. when jobs are only allowed to move at exogeneously generated time instants, a concavity property for throughput as a function of buffer parameters is established and is used to characterize an optimal allocation. Finally, a fork-join model is also treated and various further generalizations are suggested.