An iterative parallel workload balancing framework for direct condensation of substructures

This study presents a workload balancing framework which diffuses the condensation time imbalances of substructures in a homogeneous computing environment. The structure is initially partitioned in such a way that the number of substructures is equal to the number of processors. Then, the estimated condensation time imbalance of the initial substructures is adjusted by iteratively transferring nodes from the substructures with slower estimated condensation times to the substructures with faster estimated condensation times. Examples which illustrate the applicability and efficiency of this framework are presented. In these examples, the effect of utilizing different repartitioning and equation numbering algorithms are investigated.