From: Nicolas Sapay (n.sapay_at_ibcp.fr)
Date: Mon Dec 05 2005 - 08:17:58 CST
Dear Ravinder,
In my case, I write outputs on the "master" node, i.e. the first node of
the nodelist (I'm not sure it can be called "master" in fact).
Assuming that the job runs on 3 nodes: node1.myserver.edu,
node2.myserver.edu, node3.myserver.edu. The process is launched : a
unique directory is created in the </tmp> of the master node (e.g
node2.myserver.edu). Outputs will be written is this directory. However,
you will be able to see NAMD processes on each node (= 1 charmrun + 1
namd2 processes by processor). This processes consumes CPU and memory
(depend on the load balancing) but no space in the /tmp (nodes 2
excepted of course). So, you can use the /tmp of non-master nodes if it
is necessary since nothing is written on theses nodes.
Cheers
Nicolas Sapay
PS : I'm using a PBS server to run my jobs, I'm not sure this answer is
available in a different case...
Le lun 05/12/2005 à 09:09, Ravinder Abrol a écrit :
> Hi All,
>
> I am running NAMD using multiple processors and want to find out
> how the output (e.g. the dcd file) is written. More specifically,
> if only one processor writes the output how do I find out which one
> writes it and if there is a way for me specify that.
>
> The reason for this question is that since we are running NAMD on a cluster
> of processors, each of which has a temp area not readable by other processors,
> we want to see if we can utilize this temp space till the job is finished.
> The size of dcd files for >10ns simulations reaches >5GB and if we can
> use the temp space of one processoIf I have understand your question, r, that will help us greatly.
>
> Thanks,
> Ravi
>
> --
> __________________________________
>
> Ravinder Abrol
> 139-74, Beckman Institute
> Department of Chemistry, Caltech
> Pasadena, CA 91125
> Phone: 1.626.395.8143
> Fax: 1.626.585.0917
> Email: abrol_at_wag.caltech.edu
> __________________________________
-- _ Nicolas Sapay ____________________________________________ Ph.D sudent in structural bioinformatics Institut de Biologie et Chimie des Proteines CNRS - Claude Bernard University, IFR128 Bioscience Lyon-Gerland 7, Passage du Vercors Tel: +33 (0)4 72 72 26 46 69367-F Lyon cedex 07 Fax: +33 (0)4 72 72 26 04 France Web: http://pbil.ibcp.fr/
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:40:11 CST