Details

Autor(en) / Beteiligte

• Krysl, Petr
• Bittnar, Zdeněk

Titel

Parallel explicit finite element solid dynamics with domain decomposition and message passing: dual partitioning scalability

Ist Teil von

• Computers & structures, 2001, Vol.79 (3), p.345-360

Ort / Verlag

Elsevier Ltd

Erscheinungsjahr

2001

Quelle

Access via ScienceDirect (Elsevier)

Beschreibungen/Notizen

• Parallelization of explicit finite element dynamics, based on domain decomposition and message passing, may utilize one of the two partitioning cuts, namely cut led through the nodes and element edges or faces (node cut), or cut led across elements, avoiding nodes (element cut). The cost of serial explicit finite element dynamics (without considerations of mechanical contact) is almost wholly associated with elements (internal force evaluation and material updates). Sharing of nodes among processors leads to very little duplication of computing effort, and the node-cut partitioning has been used exclusively in the past. The dual nature of the element-cut partitioning and, in particular, the fact that the nodes are assigned uniquely to partitions, means that communication requirements may be in some situations quite different compared to the node-cut partitioning. Hence, the question arises whether using element-cut partitioning would make certain algorithms, such as for instance subcycling and mechanical contact, simpler, more efficient, or plainly possible. Seeking an answer to this question makes sense only if the larger overhead associated with the duplication of elements does not prevent the element-cut partitioning from being scalable as the number of processors increases, especially in fixed-partition-size situations. Here, we show that the element-cut partitioning strategy does scale, and hence presents a viable alternative to the traditional node-cut approach. We document not only the high-level algorithms but also the relevant communication code fragments of the message passing implementation using the MPI library, so as to empower the reader to fully verify our numerical experiments.