From: Gengbin Zheng (gzheng_at_ks.uiuc.edu)
Date: Wed Mar 02 2005 - 21:15:41 CST
I suspect you are running the clustermatic or Scyld version of
Charm/NAMD, in which case I decided at time of porting that Charm++
should fire NAMD jobs on available nodes without the need of nodelist file.
My theory was that Scyld/Clustermatic type of cluster normally is used
for a single job, it is not ideal for running multiple jobs due to the
contention for the master node - because of the file I/O (reading namd
config, data etc), one process of each NAMD job needs to be placed on
the master node. That is, if you are running multiple jobs, the master
node is then heavily used, which apparantly has negative impact on all
However as I realized that it is popular to use such cluster as shared
resource, I now decided to enable the feature of using nodelist to
specify the nodes for running NAMD. Also luckily, the problem of heavily
overloaded master node hopefully can be alleviated by load balancing
capability of Charm++ that migrates some of the workload away from
Please see my next email for how to use this new feature.
Parag Adhangale wrote:
>This mailing list is of great help. Thanks for all the
>I tried running NAMD using charmrun by setting up a
>node list file and using the ++nodelist option.
>Charmrun however, seems to ignore it and goes to the
>that are up. Is there a way to force charmrun to only
>use nodes specified in the nodelist file ?
>Celebrate Yahoo!'s 10th Birthday!
>Yahoo! Netrospective: 100 Moments of the Web
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:40:33 CST