A nested layered network model for parallel solutions of discrete SPPs

Abstract A nested layered netowrk mapping and algorithm is presented for parallel computational solutions of discrete stochastic programming problems with random coefficients. The algorithm is general purpose and efficiently implementable on multiprocessor systems with associative memory. Details of its implementation on a 64-CPU parallel computer and results are presented for a hard problem arising in antibody-inventory optimization in vitro. Performance is discussed along with scope vis-a-vis other general purpose efficient algorithms for discrete SPPs. Scope and Purpose: Many real world optimization problems require solutions of large scale discrete programming problems with random or uncertain coefficients and inputs. Computational difficulties arise owing to the involvement of multiple summations of large number of nonlinear mathematical terms which may not satisfy any common regularity condition within the range of summations. Parallel and associative memory based algorithms are currently in demand for solving such problems. A general purpose computing network design is presented here which contributes in the above direction. Parallel implementation and results are illustrated on a real-world problem of cost (energy) minimization.