From: Gengbin Zheng (gzheng_at_ks.uiuc.edu)
Date: Tue Dec 23 2003 - 17:22:29 CST
Hi,
We are not completely without hope... :-)
I have compiled Charm++/NAMD on both of my Laptop(Linux RH9 2GHz P4) and a
desktop (running Windows XP and Cygwin, 2.4G P4) at home and now I am
running NAMD happily on both them using 100M switch broadband router for
Apoa1 benchmark. I got about 4.52 seconds per step, while running on my
laptop Linux alone, I got 7.49s per step, so not bad - that is about
speedup of 1.66 on 2 cpus.
To do this, one needs to use /Zp4 for VC++ to pack structures in a 4 bytes
boundry (default is 8 bytes, while gcc uses 4 bytes), and needs the charmd
from the latest version with my previous fixes. I also have the Cygwin
version of NAMD2 compiled using Cygwin/g++ working too.
Here is a sample nodelist file when I start charmrun on Cygwin box.
(192.168.2.100)
group main
host 192.168.2.102 ++pathfix /home/work/ /home/gzheng/
host 192.168.2.100 ++pathfix /home/work/ C:\cygwin\home\work\
You need to map cygwin path to Windows absolute path. And always use
windows path even though you are under cygwin to charmd.
Currently I have to copy binaries to the same directory with the benchmark
config file since NAMD is not able to locate the config files across
different platforms:
./charmrun.exe +p2 ./namd2 apoa1.namd ++verbose
have fun,
(btw, regarding your PVM question, Charm++ doesnot support PVM any more.)
Gengbin
On Mon, 22 Dec 2003, Gadi Oron wrote:
> Thank you for your help.
>
>
> > My oppinion is to exclude Win** from your
> > calculations)))
>
> So the conclusion is that mixing platforms is impossible with Charm++/NAMD.
> I've tried it (just in case) and it simply does not work (RSH or charmd),
> even though IMHO there is no endian/int size issues (i386 Linux to
> Win2K+Cygwin on i386).
>
> This is a pity, since it would have allowed me to harness a lot of computer
> power for my simulation :-(
>
> BTW: Does charm++ uses PVM? I've noticed PVM files in the Charm++ directory
> but I can't figure out where it is used.
>
> Again, many thanks for your efforts.
>
> -Gadi
>
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:37:15 CST