From: David A. Horita (dhorita_at_wfubmc.edu)
Date: Fri Jul 10 2009 - 13:37:54 CDT
Regarding scaling, what charmrun options are you using? +giga and
+strategy USE_GRID/USE_MESH make a big difference for me. Also, how
many cores per box does each system have (I guess most new ones are at
8, but if it's different, please note that). Once you're computing
between boxes, the network comes into play (so also please specify
gigabit, infiniband, etc.) I get very good scaling throughout an
infiniband-linked group, decent scaling throughout a gigabit-linked
group, and miserable scaling between groups (slower network with lots of
other traffic on it). NAMD with PME pushes a lot of data, so there are
numerous factors that contribute to scaling.
Thanks,
David
-----------------------------
David A. Horita, Ph.D.
Department of Biochemistry
Wake Forest University School of Medicine
Winston-Salem, NC 27157-1016
Tel: 336 713-4194
Fax: 336 716-7671
email: dhorita_at_wfubmc.edu
web: http://www1.wfubmc.edu/biochem/faculty/Horita.htm
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:53:01 CST