Simulation of particle mixing by turbulent convective flows on the Connection Machine

Abstract

Mixing of particles by chaotic flow fields was simulated on the Connection Machine. We assigned each cell to the processor and kept the coordinates of particles residing on the cell in the local memory of the processor. This approach implies the exchange between the local memories, when a particle moves from one cell to another. Approximately 1 d particles were injected into atime-dependent flow f e l d obtained by solving the nonlinear system of PDEs, describing turbulent thermal convection. The flow field was calculated on CRAY and data were transferred to CM-200 through high-speed HlPPI channel.

Experiments with a gigabit neuroscience application on the CM-2

“…The application is a neural simulation program that runs on the Thinking Machines CM-2; the program generates new data at a rate of To date, there is little published literature on distributing computation among supercomputers connected together by high speed networks. Although earlier work of Mechoso et aZ [8] and Malevsky et al [7] have explored the decomposition of applications among supercomputers, little is known about the interaction of the machines and the applications that require the transport of very high speed data streams. Previous high speed network tests performed by Clinger [4] and Beach [1] did not include the CM-2 and the Convex C3880, the machines we are using for our experiments.…”

Interactive interrogation of computational mixing data in a virtual environment

“…For large unsteady datasets, this can be problematic. Another option, explored by Malevsky, et al (Malevsky et al, 1992) is to distribute the grid evenly between the processors and pass the particles between the processors as the particles move from one cell to the next. However, the performance of this method suffers if the particles are not evenly distributed over the grid.…”