[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Fwd: Re: [ccp4bb]: Parallelization of ccp4 programs on linuxcluster ?]
*** For details on how to be removed from this list visit the ***
*** CCP4 home page http://www.ccp4.ac.uk ***
-----Forwarded Message-----
> From: Michel Fodje <michel.fodje@mbfys.lu.se>
> To: Paul Emsley <paule@chem.gla.ac.uk>
> Subject: Re: [ccp4bb]: Parallelization of ccp4 programs on linux cluster ?
> Date: 29 Aug 2002 10:31:30 +0200
>
> There are several parallel implementations of the FFT routine. Since
> many CCP4 programs do use FFT, it could be trivial to have these
> programs call the parallel FFT instead thus achieving widespread
> parallelization of most programs. This is already possible with CNS on
> SGI by using SGI's parallel FFT routines.
>
> There is a GPL parallel FFT library called FFTW developed at MIT which
> is used in GROMACS 'The world's fastest molecular dynamics'. It performs
> quite well through LAM-MPI (http://www.lam-mpi.org) or MPICH
> (http://www-unix.mcs.anl.gov/mpi/mpich/).
> I'm not sure how the licensing is compatible with CCP4. It is primarily
> a C-library but Fortran bindings exist. Converting a program which uses
> FFTW to be parallel is as easy as renaming the subroutine within the
> code.
>
> /michel
>
> On Wed, 2002-08-28 at 17:10, Paul Emsley wrote:
> > *** For details on how to be removed from this list visit the ***
> > *** CCP4 home page http://www.ccp4.ac.uk ***
> >
> > On Tue, 2002-08-27 at 05:36, Martin Laurberg wrote:
> > > *** For details on how to be removed from this list visit the ***
> > > *** CCP4 home page http://www.ccp4.ac.uk ***
> > >
> > > Dear all,
> > > I have a small cluster of Linux Redhat7.3 openMosix machines
> > > and would like to take advantage of program parallelization.
> > > Has anyone had success on parallelizing parts of the ccp4
> > > suite under linux ?
> > >
> >
> > It is my understanding that this has considered by those involved - in
> > fact I may be so bold as to suggest that there is some enthusiasm for
> > the idea.
> >
> > It is much easier to implement parallelization at the shell-script
> > (ccp4i?) level rather than the fortran level - and this will be
> > attempted first (IIUC).
> >
> > As you know, CCP4 software outputs files as the results of the
> > calculations. Thus there is a burden on the operating system to handle
> > some of the adminstrative load and (consistenly used) automount and
> > ssh (or a network queueing system) will be required (IIUC, you
> > understand). You might like to voice your opinion on the
> > acceptability of this.
> >
> > If you ask this question again in 6 months time, you may get a more
> > fully-rounded answer.
> >
> > HTH somewhat,
> >
> > Paul.
> >
> >
> >
> --
> _______________________________
> Michel N. Fodje
> Molecular Biophysics,
> Lund University, Sweden
>
> phone: +46 46 222 45 13
> email: michel.fodje@mbfys.lu.se
> web: http://www.mbfys.lu.se
> _______________________________
--
_______________________________
Michel N. Fodje
Molecular Biophysics,
Lund University, Sweden
phone: +46 46 222 45 13
email: michel.fodje@mbfys.lu.se
web: http://www.mbfys.lu.se
_______________________________