We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.DC

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Distributed, Parallel, and Cluster Computing

Title: CROFT: A scalable three-dimensional parallel Fast Fourier Transform (FFT) implementation for High Performance Clusters

Abstract: The FFT of three dimensional (3D) input data is an important computational kernel of numerical simulations and is widely used in High Performance Computing (HPC) codes running on large number of processors. Although the efficient parallelization of 3D FFT has been largely investigated over the last few decades, performance and scalability of parallel 3D FFT methods on new generation hardware architecture for HPC is a major challenge. Looking at upcoming exascale cluster architectures, the conventional parallel 3D FFT calculations on HPC needs improvement for better performance. In this paper, we present CDACs three dimensional Fast Fourier Transform (CROFT) library which implements three dimensional parallel FFT using pencil decomposition. To exploit the multithreading capabilities of hardware without affecting performance, CROFT is designed to use hybrid programming model of OpenMP and MPI. CROFT implementation has a feature of overlapping compute and memory I/O with MPI communication. Depending on the number of processes used, CROFT shows performance improvement of about 51 to 42 percent as compared to FFTW3 library.
Comments: 28 Pages, 15 Figures
Subjects: Distributed, Parallel, and Cluster Computing (cs.DC); Performance (cs.PF)
Cite as: arXiv:2002.04896 [cs.DC]
  (or arXiv:2002.04896v1 [cs.DC] for this version)

Submission history

From: Vivek Gavane [view email]
[v1] Wed, 12 Feb 2020 10:12:15 GMT (4531kb)

Link back to: arXiv, form interface, contact.