Parallel processing of excavation in soils with randomly generated material properties

Today, affordable high performance clusters enable engineers to carry out large nonlinear 3D analyses very quickly using parallel processing. It has been estimated that by 2018, a 1 Petaflop cluster (with 100,000 cores) will cost around $150,000; a price affordable by a reasonably sized engineering firm or University department. With these significant improvements in technology, it is becoming increasingly cost-effective to incorporate uncertainty into analyses, for example by undertaking Monte Carlo simulations with randomly generated soil properties. As each realisation is independent, it can be executed at the same time. For large models that require move memory than is available on a single core, each realisation can be solved by subdividing the problem over multiple cores. The authors will present results for an excavation problem using this two-level parallelisation strategy. The parallel software used, ParaFEM, has been recently updated to interface with a number of external tools including the RFEM library, the visualisation tool ParaView and the graph partitioner METIS. For a single realisation, ParaFEM can make good use of 32,000 cores and solve problems with 1 billion degrees of freedom.