From: Ajasja Ljubetič (ajasja.ljubetic_at_gmail.com)
Date: Wed Jul 28 2010 - 04:42:20 CDT
I think it is a shame that the later versions of NAMD (I think after 2.5)
are not supported on multiple windows nodes.
Especially since I have 8 nodes each with 8 processors running 32 bit winxp
connected using gigabit LAN:)
I have tried using charmrun to run a simulation on many nodes. I downloaded
the charm-6.2.1_net-win32_production binaries and ran the charmd on two
.\charmrun.exe C:\NAMD2.73b\namd2.exe L35_WAT_min_fix1.conf ++nodelist
nodelist ++p 16 > test.log
I am able to get charmrun to run 8 copies of namd2.exe (by copying the namd
and input files to the same path on both nodes)
charmd reports something like
Listening for requests on port 12396
Connection from IP 220.127.116.11, port 49174 at Wed Jul 28 11:35:57 2010
and argLine ' L35_WAT_min_fix1.conf'
and environment 'NETSTART=14 18.104.22.168 65532 1488 0'
The processor usage on both nodes goes to 100%, but nothing gets writen to
the output. The log file test.log contains
CmiMemory: fences and atomic operations not available in native assembly
 isomalloc.c> Disabling isomalloc because mmap() does not work
Does anyone know where the problem is? Do I need to compile NAMD myself to
support charmrun? (and will it suffice to change
CHARMARCH = multicore-win32 to CHARMARCH = net-win32 in the make files?)
And is there a good reason that NAMD on multiple windows nodes
not supported any more?
I also have sshd installed on each node. Is it better to use charmd or sshd
or mpich2 (maybe it makes no difference on an internal network)?
Any help is greatly appreciated.
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:54:21 CST