Scalability analysis of partitioning strategies for finite element graphs: a summary of results

The authors present a scalability analysis of three partitioning strategies, namely striped partitioning, binary decomposition, and scattered decomposition. The analysis is performed using the Isoefficiency metric, which helps in predicting the performances of these schemes on a range of processors and architectures. The performance of each of these schemes is related to the various problem characteristics such as mesh geometry and density. Isoefficiencies are presented for hypercube and mesh connected architectures. Theoretical results are verified through simulations.< >