This benchmark is similar to Bench_BSSN_PUGH, but it uses mesh refinement instead of a uniform grid. It also uses the driver Carpet instead of PUGH. You find more information about Carpet on its web page. Carpet is still under development. It scales up to about 100 processors.

This benchmark comes in two configurations, 1lev and 3lev. The first is a unigrid configuration, which should in an ideal world show the same performance characteristics as PUGH. The second is a mesh refinement configuration, which requires more commBunication for the same number of grid points. In return, there is more spatial and temporal resolution. You should run both configurations.

Each configuration comes in two sizes: 1lev_60l and 3lev_32l use about 400 MB, and 1lev_75l and 3lev_42l use about 800 MB of main memory per CPU. You should run both sizes of each configuration if you have enough memory.

For the interested reader, we make the two parameter files and the thorn list here available.

• Bench_BSSN_Carpet_1lev_60l.par
• Bench_BSSN_Carpet_1lev_75l.par
• Bench_BSSN_Carpet_3lev_32l.par
• Bench_BSSN_Carpet_3lev_42l.par
• Bench_BSSN_Carpet.thornlist

These files are also contained in the benchmark's tarball.