(298d) Optimizing Production Planning Using High Performance Computing

Varvarezos, D. - Presenter, Aspen Technology, Inc.

Historically new hardware has delivered significant performance improvements while reducing computing hardware costs. In the last few years, however, issues of power consumption and heat generation are limiting further performance increases. The new emerging trend favors parallelism in the form of increasing numbers of execution cores per CPU chip. Multi-core, cluster, and cloud computing have now entered the mainstream application space. Presently, all important enterprise planning solutions will not "automatically" run faster on multi-core computers. This paper addresses this issue and presents a comprehensive approach to utilizing the emerging multi-core architectures (as well as high performance computing at large) in production planning. The proposed methodology is completely scalable from multi-core computers, to multi-node clusters and it is based on coarse-granular parallelization utilizing the MPI standard. A custom load-balancing algorithm further optimizes this parallelization process. Solution times for actual production planning models using this technology are scaling almost linearly to the number of processor cores deployed. This methodology was successfully applied and validated on several computing platforms ranging from 2-core and 8-core computers to 40-core clusters. The results of this work demonstrate significant business benefits from (a) allowing planners to react quickly to an ever-changing business environment by identifying and responding quickly to opportunities (through solving more case scenarios faster), and (b) enabling the practical achievement of (stochastic) globally optimal solutions as part of the regular planning work process.