Unsteady Turbomachinery Computations Using Massively Parallel Platforms

Abstract

The CPU power offered by the latest generation of supercomputers is such that the unsteady simulation of the full wheel of a compressor or turbine is now within reach. This CPU power is mainly obtained by increasing the number of processors beyond several thousands, e.g. the BlueGene/L computer of Lawrence Livermore National Laboratory has approximately 130,000 processors. Consequently extreme care must be taken when the simulation codes are ported to these platforms. This paper discusses the computer scientific aspects of simulating unsteady turbomachinery flows on massively parallel systems when multi-block structured grids are used. Load balance, parallel IO and search algorithms are addressed, especially for the case where the number of processors is larger than the number of blocks, i.e. when blocks must be split during runtime. Preliminary results for cases with more than 200 million nodes running on 1,800 processors are presented.