Keywords: multi-scale, distributed memory, parallel computing, scalability, aero-acoustics, coupled simulation.

The focus in this work is put on the scalable deployment of the overall scheme for highly distributed parallel computing systems. It is important for any computational intensive simulation to provide numerical strategies which are fit for today's supercomputing facilities in order to make efficient use of them. A very important feature for those parallel machines with distributed memory, is the scalability. The scalability in turn can be viewed as a successful separation of independent tasks. In the described coupling mechanism the computation is split up naturally by the various domains considered in the simulation and artificially by mesh partitioning within these domains. For the maximal possible scalability all communications and interdependencies are expressed on the processor specific level, that is between partitions of the various domains which we call sections.

But not only the scaling in the distribution of the computational load, that is computing cycles or actual run-time needs to be satisfied for a successful deployment of large scale simulations. Another important feature, that needs to be considered in terms of scalability is the memory consumption, where a single peak of consumed memory at any time on any of the processes can disable the complete simulation of the distributed memory machine. In this paper we describe a completely distributed approach for the mesh representation, which allows the easy parallel handling of various parts in the domain with minimal overhead during the parallel computation. For this we describe a general elemental unstructured mesh format, which makes use of space-filling curves and defines an easily read file representation for efficient loading of elemental data. The neighbour information are stored for each element, which allows for independent reading of each element. In total it is shown, that a scalable implementation is possible for complex simulation setups.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £85 +P&P)