From: Gengbin Zheng (gzheng_at_ks.uiuc.edu)
Date: Wed Mar 02 2005 - 21:29:40 CST
Hi Rene,
As a matter of fact, charmrun in Scyld/Clustermatic version did not
take any nodelist file, that is even if you specify one, it is silently
ignored.
Per popular request, I now implemented this feature for Scyld and
Clustermatic version, so that it can take a nodelist file.
I have implemented this in the latest charm in CVS, one could checkout
the lastest Charm from cvs:
http://charm.cs.uiuc.edu/download/
or wait until tomorrow (after 2am CST) to download the nightly tested
version of charm at:
http://charm.cs.uiuc.edu/autobuild/cur/
please download the source code, as we don't have pre-compiled binary
version for clustermatic version.
Once you checkout charm and compile it with:
./build charm++ net-linux clustermatic, or
./build charm++ net-linux scyld
the new charmrun should have the feature of taking a nodelist file as
an option:
./charmrun ++nodelist ./nodelist ./namd2 config
Note that you MUST use ++nodelist option, otherwise, it will act same as
before by choosing compute nodes itself.
The nodelist file should contain compute node number like:
group main
host -1
host 0
host 5
host 8
NOTE that if the file system is not NFS mounted, you need to use "-1"
the master node as the first line of host. Otherwise, NAMD won't be able
to read config and data file if it is running on slave nodes.
Hope this helps and please let me know if you have any problem,
Gengbin
Rene Salmon wrote:
>Hi list,
>
>Can some one point me in the right direction as to how to get charm and namd
>to use a hostlist file?
>
>
>The problem is that in our cluster sometimes we get assigned nodes that are
>not in sequential order. For Example you might get assigned nodes (2 3 4
>6 9 11)
>
>The syntax we use to run charm and namd on a clustermatic cluster is as
>follows:
>
>charmrun ++skipmaster ++verbose ++startpe $first_node ++endpe $last_node
>++ppn 2 ++p $num_procs namd2 alanin
>
>where startpe would get node 2 and endpe would get node 11. The problem is
>that charm then tries to sequentially contact all nodes starting at node 2
>up to node 11. is ther any way to get charm to only contact the nodes in a
>nodelist file?
>
>we tried this as the node list file:
>
>host 2
>host 3
>host 4
>host 6
>host 9
>host 11
>
>
>with this command:
>
>charmrun ++skipmaster ++verbose ++nodelist nodelistfile.txt ++startpe
>$first_node ++endpe $last_node ++ppn 2 ++p $num_procs namd2 alanin
>
>but this does not work.
>
>Thank you for any help on this.
>
>Rene
>
>
>
>
>
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:39:11 CST