This plot compares the performance and scaling of the code on various machines. The Machines are : IBM SP3 (Seaborg) 16 way 256 nodes , HP Alphaserver (QSC) using 4 way 128 nodes, MCR Linux Cluster 2 way 1024 nodes, JAZZ Linux cluster 1 way 350 nodes,
The impact of using one processor per node versus both of them on MCR
Scaling of the multigrid algorithm in FLASH. This algorithm needs much more global communication and has a much worse surface to volume ratio than most of the rest of FLASH. The first plot is on MCR, and the second one is on Seaborg.
