[molpro-user] DF-DFT-SAPT calculation requires too much memory
MayAJ1 at cardiff.ac.uk
Tue Jul 12 13:47:33 BST 2011
This job should not require anywhere near that amount of memory. What
happens if you run the example without the 'memory' directive you added?
Please find attached the output of this job on our machines, which took
under 1 minute with 8 processes.
How many cores are there on the node where you ran the 8 processes?
There should be no MPI communication issues since the job was run on
only 1 node. Perhaps there are problems with the disk I/O, are any of
the disks (/scratch and /u) lustre/NFS mounted?
I also notice this is PL1, I suggest to install PL23 to get the latest
updates and bug fixes.
On 05/07/11 20:18, WU Yanbin wrote:
> Hi, everyone,
> I'm trying to do DF-DFT-SAPT calculations. And I started with the sample
> example of Ne-Ar interactions.
> The calculation was ran with 8 processors.
> The job required 490GB memory. If the allocated memory is smaller than
> 400GB, the job would be killed due to inadequate memory.
> The simulation time for the job is around one hour, which is much long
> than DFT-SAPT calculation of the same system (which only takes about 1
> minute with 500MB memory).
> The computation resources I'm using is a distributed-memory system. And
> I would like to provide more details of the clusters if necessary.
> Despite the long simulation time, the simulation result is the same as
> the sample output.
> The input and output file are attached.
> Is there any keyword I should use to reduce the memory use and
> simulation time?
> If I'm speculating that the density fitting code is not compiled
> properly, anything I should do to verify that and how I can fix this
> compiling problem?
> Any hint is appreciated. Do let me know if the question is not made clear.
> Thank you.
> Molpro-user mailing list
> Molpro-user at molpro.net
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 11367 bytes
Desc: not available
More information about the Molpro-user