Computational assessment of distributed decomposition methods for stochastic linear programs

Incorporating uncertainty in optimization models gives rise to large, structured mathematical programs. Decomposition procedures are well-suited for parallelization, thus providing a promising venue for solving large stochastic programs arising in diverse practical applications. This paper presents an adaptation of decomposition methods for execution on distributed computing systems. A regularized decomposition, as well as the linear decomposition algorithm, are implemented for execution on distributed multiprocessors. Computational results on an IBM SP2 multiprocessor system are reported to demonstrate the comparative performance of the methods on a number of test cases.