From 5b0167d759a498868f9c59b5926e6a3605b20277 Mon Sep 17 00:00:00 2001 From: Rene Gassmoeller Date: Mon, 12 Jun 2023 11:45:17 -0600 Subject: [PATCH] Remove lines that I didnt want to commit. --- README.md | 2 -- 1 file changed, 2 deletions(-) diff --git a/README.md b/README.md index cdccb245..03df6bc4 100644 --- a/README.md +++ b/README.md @@ -18,8 +18,6 @@ community and keep them in the [CONTRIBUTING.md](CONTRIBUTING.md) file in Rayleigh's repository. - - Parallelization ----------------------------- The pseudo-spectral nature of Rayleigh means that its parallelization necessarily relies heavily on *global communication* patterns. That said, Rayleigh's parallelization is based around a 2-D domain decomposition and large-message-size all-to-alls. These features allow the code to overcome many of the obstacles that traditionally limit the scalability of spectral methods. The end result is a pseudo-spectral code optimized for petascale machines. Rayleigh's pure-MPI mode has demonstrated highly efficient strong scaling on 131,000 cores of the Mira Blue Gene/Q supercomputer for problems with approximately 2048^3 grid points (2048 spherical harmonics). Performance numbers from Mira are shown below. A summary of Rayleigh's performance and how it compares against other popular dynamo codes (albeit at at smaller process counts) may be found in the recent performance benchmark results of [Matsui et al. (2016)](http://onlinelibrary.wiley.com/doi/10.1002/2015GC006159/full).