aboutsummaryrefslogtreecommitdiff
path: root/doc
diff options
context:
space:
mode:
authorschnetter <schnetter@b61c5cb5-eaca-4651-9a7a-d64986f99364>2005-09-26 03:06:25 +0000
committerschnetter <schnetter@b61c5cb5-eaca-4651-9a7a-d64986f99364>2005-09-26 03:06:25 +0000
commit381b759f8afba04cf673cb0540c142f117b84913 (patch)
tree85739e0298e574a110af421f69c4a53d9529a98b /doc
parent710e1e23cfecc96153ac103f13e04ceec625dd44 (diff)
Apply Frank Loeffler's patch to improve the processor distribution of
PUGH: Up to now, PUGH does not take the actual grid size into account while setting the number of processors/domains in each direction. This gives problems, e.g. using boxes like 200x7x7 (with 3 ghostzones). It also does not really do a good job in situations with non near-cubic domains (like in bitant mode it devides first in the z-direction, which is usually the smallest). I inserted another decomposition algorithm which is looking for the longest direction and devides this if possible. Because it was done to be easy to understand, it can fail in certain situations, in which it is gracefully falling back to the old decomposition (e.g. a cubic box and 9 processors). To include this algorithm I added a keyword "automatic_old" to PUGH::processor_topology. The new behaviour is the default; this keyword re-enables the old behaviour. git-svn-id: http://svn.cactuscode.org/arrangements/CactusPUGH/PUGH/trunk@471 b61c5cb5-eaca-4651-9a7a-d64986f99364
Diffstat (limited to 'doc')
-rw-r--r--doc/documentation.tex9
1 files changed, 9 insertions, 0 deletions
diff --git a/doc/documentation.tex b/doc/documentation.tex
index 4f9abac..53aed8d 100644
--- a/doc/documentation.tex
+++ b/doc/documentation.tex
@@ -166,6 +166,15 @@ the partition parameters) will apply the automatic distribution. That's why
it is not necessary to set \verb|PUGH::partition_3d_y = "30"| or
\verb|PUGH::partition_3d_z = "15 15"| in the parameter file.
+Because the previous automatic distribution gave problems in some
+cases (e.g.\ very long box in one, but short in other directions),
+there is now an improved algorithm that tries to do a better job in
+decomposing the grid evenly to the processors. However, it can fail
+in certain situations, in which it is gracefully falling back to the
+previous (\verb|"automatic_old"|) giving a warning. Note that, if one
+or more of the parameters \verb|PUGH::processor_topology_3d_*| or
+\verb|PUGH::partition_3d_*| are set, this mode automatically falls back to
+\verb|"automatic_old"| without warning.
\section{Understanding PUGH Output}