I reviewed that doc thank you, I wonder if it is a little out of date, mpirun redirects to mpiexec which does not support options "-nimage" and "-npool" , those are mentioned for Plane-wave tasks on page10.
I am running it now in blocks of 4 cpus to match the number-combination requirement. 4 cores on the same system completes in 1h47m , 16 cores on the same system completes in 2h19m. Is this not going to get faster on a single system until we add more memory to the system? or perhaps to drop the number of cpus specified in mpiMachinefile.txt so it spreads a fewer number of jobs over a greater number of systems? or is the bottleneck to do with the local hard drives not being fast enough? I am not familiar with the maths equations/terminology mentioned in the document. thanks, nick > -----Original Message----- > From: giannozz at democritos.it > Sent: Mon, 31 Jan 2011 11:16:06 +0100 > To: pw_forum at pwscf.org > Subject: Re: [Pw_forum] QE and mpich2, Linux > > in order to take advantage of parallelization, some understanding of > how parallelization works in QE is needed. The user guide and this > document: > http://www.fisica.uniud.it/~giannozz/QE-Tutorial/tutorial_para.pdf > contain some info. Throwing in more processors will not by itself do > the job. > > P. > --- > Paolo Giannozzi, Dept of Chemistry&Physics&Environment, > Univ. Udine, via delle Scienze 208, 33100 Udine, Italy > Phone +39-0432-558216, fax +39-0432-558222 > > > > > _______________________________________________ > Pw_forum mailing list > Pw_forum at pwscf.org > http://www.democritos.it/mailman/listinfo/pw_forum ____________________________________________________________ Send your photos by email in seconds... TRY FREE IM TOOLPACK at http://www.imtoolpack.com/default.aspx?rc=if3 Works in all emails, instant messengers, blogs, forums and social networks.
