Efficiently Using Memory in Lattice Boltzmann Simulations

Numerical simulation programs using the lattice Boltzmann equation are limited in the range of problems they can address by memory requirements. We describe an implementation scheme that can reduce memory usage by up to 78%. The performance of both the simple and the memory-reduced implementation are compared. Memory access is the determining factor in the speed of execution of memory intensive applications. We therefore look at the cache memory utilization in both the storage reduced version and in the simple reference implementation. From this we conclude that the memory-reduced method does not degrade the performance, in terms of execution speed, relative to a simple implementation.