From a10a5ba686830a2947d0235cb5ce2fb22bc14f4d Mon Sep 17 00:00:00 2001 From: Erik Schnetter Date: Mon, 14 Jan 2008 14:50:00 +0000 Subject: CarpetWeb: Add scaling results from Ranger darcs-hash:20080114145039-dae7b-5c7a54e9c35863199f9faaa99b32325f249520cd.gz --- Carpet/CarpetWeb/index.html | 37 +- Carpet/CarpetWeb/scaling-ranger/Makefile | 17 + Carpet/CarpetWeb/scaling-ranger/findrow | 54 +++ Carpet/CarpetWeb/scaling-ranger/loopprocs | 11 + .../scaling-ranger/results-ranger.gnuplot | 35 ++ Carpet/CarpetWeb/scaling-ranger/results-ranger.out | 474 +++++++++++++++++++++ Carpet/CarpetWeb/scaling-ranger/results-ranger.pdf | Bin 0 -> 4272 bytes Carpet/CarpetWeb/scaling-ranger/results-ranger.png | Bin 0 -> 8732 bytes 8 files changed, 627 insertions(+), 1 deletion(-) create mode 100644 Carpet/CarpetWeb/scaling-ranger/Makefile create mode 100755 Carpet/CarpetWeb/scaling-ranger/findrow create mode 100755 Carpet/CarpetWeb/scaling-ranger/loopprocs create mode 100644 Carpet/CarpetWeb/scaling-ranger/results-ranger.gnuplot create mode 100644 Carpet/CarpetWeb/scaling-ranger/results-ranger.out create mode 100644 Carpet/CarpetWeb/scaling-ranger/results-ranger.pdf create mode 100644 Carpet/CarpetWeb/scaling-ranger/results-ranger.png (limited to 'Carpet/CarpetWeb') diff --git a/Carpet/CarpetWeb/index.html b/Carpet/CarpetWeb/index.html index 9e4d556a4..3de34d4e9 100644 --- a/Carpet/CarpetWeb/index.html +++ b/Carpet/CarpetWeb/index.html @@ -107,6 +107,41 @@

News

+
+

January 14, 2008: Carpet's communication + infrastructure has been improved significantly, making Carpet + scale to at least 4,000 processors, including mesh refinement. + Using "friendly user time" + on Ranger, + the new 60,000 + core TeraGrid + supercomputer + at TACC, we measured + the benchmark results below for a numerical relativity kernel + solving the BSSN equations. These benchmarks emply a hybrid + communication scheme + combining MPI + and + OpenMP, using the shared + memory capabilities of Ranger's nodes to reduce the memory + overhead of parallelisation. We are grateful for the help we + received from Ranger's support team.

+

The graph below shows weak scaling tests for both unigrid and + mesh refinement benchmarks. The problem size per core was + kept fixed, and there were 4 OpenMP threads per MPI process, + with 1 MPI process per socket. The benchmark was also run + with the PUGH driver for comparison for certain core counts. + As the graphs show, this benchmark scales near perfectly for + unigrid, and has only small variations in run time for nine + levels of mesh refinement.

+
+

Scaling graph for Ranger

+
+