The problem of scheduling customer orders in the management of a large multi-product warehouse is addressed under realistic assumptions including rapidly changing data, uncertainties and unforeseen events. Simple flexible solution strategies are sought, in the sense that the scheduling problem should not be solved again from the beginning everytime new information becomes available. Among the solution strategies, one is based on a few statistical assumptions about the random variables involved in the process: it exploits a stochastic version of the well-known duality principle for nonlinear programming and yields a number of interesting features.

The evolution of Computer Integrated Manufacturing calls for the organizatlon of complex production processes, obtained by the composition of automated job shops. The larye-scale event-driven nature of such systems requires a revision of the techniques for designing production planning and control architectures. Two manufacturing environments are here considered : batch and repetitive. In the former case, productlon batches are closely related to customer orders so that th eplanning problem consists of scheduling and sequencing the lots through the sequence of shops in order to meet due dates. In the latter case, stationary or slowly varying order quantities must be produced. Then production rates and lot sizes are the main control variables. The paper shows a unified approach to the production planning problem for batch and repetitive manufacturing. Starting from a common model and control architecture, specialized algorithms are derived for the two cases. A hierarchical and decentralized architecture is stressed.

The performance of aa delta interconnection network for multiprocessors is evaluated in a circuit switching environment An error is pointed out in previous literature and an exact analytical model is given for regeneration systems, where a connection request is considered lost if not immediately granted. An approximated numerical method is suggested for the correction of the analytical results, which gave outputs in very good agreement with the simulation of real systems where requests are maintained.

The performances of a Flexible Manufacturing System with multiple part classes are evaluated by closed queueing network analysis. The problem considered here refers specifically to the presence into the system of a real time intelligent router. It is well known that this condition sharply contrasts with the classical hypothesis needed for an exact solution of the steady state condition of the network (product form solution). Heuristic load-dependent coefficients have been introduced in the Reiser's recursive algorithm of Mean Value Analysis to model a dispatcher that routes parts to a set of equivalent machines according to the unfinished workload. The method has been extended to the case of unequal machine loads. Numerical results are given to show the achievable accurancy of the approximation.