FOSS4G-Asia 2023 Seoul

Memory-Efficient Flow Accumulation Using OpenMP Parallelization
11-30, 14:30–14:50 (Asia/Seoul), Seoul Archive

This talk introduces a new open-source fast Memory-Efficient Flow Accumulation (MEFA) algorithm using OpenMP parallelization. Flow accumulation is one of the most important parameters for any hydrologic and/or environmental analyses. There have been many studies that focus on the improvement of its computational efficiency using OpenMP, MPI, and CUDA, but most of benchmark algorithms are not memory-efficient in terms of the number of write operations and the amount of bytes read/written. This memory-inefficient nature of those algorithms deteriorates their computational performance during parallel processing of the input flow direction matrix. The new MEFA algorithm has reduced its memory requirements in both writes and bytes by eliminating a need for intermediate output matrices and writing its final value to each cell only once deterministically. The new algorithm performed 45% and 19% faster than its OpenMP and MPI benchmark algorithms, respectively, using 20% less memory asymptotically.

Huidae /hidɛ/ Cho is a member of the GRASS GIS Development Team and Project Steering Committee. He researches and teaches Water Resources Engineering in the Department of Civil Engineering at New Mexico State University.

This speaker also appears in: