[molpro-user] How to run molpro in parallel?

[Shenggang Li]

Thanks, Reuti, I will ask the vendor about this.

On 3/18/06, Reuti <reuti at staff.uni-marburg.de> wrote:
> Quoting Shenggang Li <shenggangli at gmail.com>:
>
> > I tried to run in serial, 2 processors on a single node, or 2
> > processors on two nodes, but it ended with the same error message.
> > The hostname shouldn't be a problem, as I have sucessfully run NWChem,
> > as well as molpro build without infiniband support.  What is the
> > registry file anyway?  Thanks for your help!
>
> Google told me, that the registry is an Infiniband thing. So I'd
> suggest to ask
> on any Infiniband list or vendor.
>
> -- Reuti
>
> >
> > On 3/17/06, Reuti <reuti at staff.uni-marburg.de> wrote:
> >> Quoting Shenggang Li <shenggangli at gmail.com>:
> >>
> >> > Thanks a lot!  That seems to work.  However, I got the following error
> >> > message, when I tried to tune the package (bin/molpro -n 1
> >> > --mpirun-machinefile /home/sli/hostfile tuning.com &):
> >> >
> >>
> >> Did you tried with 2 or more CPUs/nodes in parallel?
> >>
> >> > MPP nodes  nproc
> >> > compute-1-1.loca    1
> >> > ga_uses_ma=true, calling ma_init with heap size          10000
> >> > determined by size of -G / nprocs
> >> >
> >> > Primary working directories:    /home/sli/scratch/molpro
> >> > Secondary working directories:  /home/sli/scratch/molpro
> >> >
> >> > blaslib=default
> >> >
> >> > MPP tuning parameters: Latency=  1000 Microseconds,   Broadcast
> >> > speed=   20 MB/sec
> >> > default implementation of scratch files=df
> >> >
> >> > ? Error
> >> > ? Failure to open registry file variable.registry
> >> > ? The problem occurs in open_registry_file
> >> >
> >> > ERROR EXIT
> >> > CURRENT STACK:      MAIN
> >> > ? Error
> >> > ? Variable STATUS not registered
> >> > ? The problem occurs in setvar
> >> >
> >> > ERROR EXIT
> >> > CURRENT STACK:      MAIN
> >> >
> >> > RECURSIVE CALL OF ERROR HANDLER, FILE INFORMATION MAY BE LOST
> >> >
> >> > EMERGENCY STOP
> >> >
> >> > GA ERROR termxy
> >> >
> >> > GA ERROR ^!~V*^@^@^@
> >> > 0:0:termxy:: 0
> >> >
> >> > Does this mean my compiled binaries are bad?  What is the
> >> > variable.registry file and where should it be?  The hostname is
> >> > compute-1-1.local and it was printed out as compute-1-1.loca.  Is the
> >> > hostname too long?
> >>
> >> Interesting - counts just 16. This would be a limitation of GA. I'm
> >> not sure,
> >> whether it would help to put only the short alias name in the hostfile and
> >> additionally in /etc/hosts. Might give problems, as AFAIK Rocks
> >> gives also the
> >> domainname in the hostname command.
> >>
> >> -- Reuti
> >>
> >> >
> >> > Regards,
> >> >
> >> > On 3/17/06, Reuti <reuti at staff.uni-marburg.de> wrote:
> >> >> Quoting Shenggang Li <shenggangli at gmail.com>:
> >> >>
> >> >> > Thanks for your reply.  It won't work though.  mpirun_ssh requires
> >> >> > -hostfile <hostfile> option, which cannot be replaced by the
> >> >> > -machinefile <machinefile> optioin.  Do we have to modify the code to
> >> >> > change the molpro file?
> >> >>
> >> >> You could of course change the name in the Molpro source, but I'd go
> >> >> for a small
> >> >> wrapper around mpirun_ssh to change the optionname, something like:
> >> >>
> >> >> #!/bin/sh
> >> >> exec mpirun_ssh.orig "${@/machinefile/hostfile}"
> >> >>
> >> >> -- Reuti
> >> >>
> >> >> >
> >> >> > On 3/16/06, Reuti <reuti at staff.uni-marburg.de> wrote:
> >> >> >> Hi,
> >> >> >>
> >> >> >> you can give to the molpro command:
> >> >> >>
> >> >> >> molprop -n <slots> --mpirun-machinefile <machinefile> < <infile>
> >> >> > <outfile>
> >> >> >>
> >> >> >> which should provide the usual argument to the mpirun command
> >> when it's
> >> >> >> called.
> >> >> >>
> >> >> >> HTH - Reuti
> >> >> >>
> >> >> >>
> >> >> >> Quoting Shenggang Li <shenggangli at gmail.com>:
> >> >> >>
> >> >> >> > Dear molpro users,
> >> >> >> >
> >> >> >> > I have managed to build MOLPRO 2002.6 on a Xeon em64t machine with
> >> >> >> > Infiniband support.  However, I can't run it.  The problem is due to
> >> >> >> > its parallel wrapper, mpirun_ssh, which takes the following command
> >> >> >> > line arguments:
> >> >> >> >
> >> >> >> > mpirun_ssh ... -np <n> -hostfile <hostfile> <executable> <arguments>
> >> >> >> >
> >> >> >> > However, the molpro script doesn't provide the "-hostfile
> >> <hostfile>"
> >> >> >> > part and the execution failed.  I tried to capture the command
> >> >> >> > arguments passed to mpirun_ssh by replacing it with a script to echo
> >> >> >> > the arguments and I got
> >> >> >> > -np <n> <molpro_executable>.
> >> >> >> > Then I use the following script to pass the arguments to mpirun_ssh:
> >> >> >> > <mpirun_ssh> -np $2 -hostfile <hostfile> <molpro_executable>
> >> >> >> > but it still failed to run.
> >> >> >> >
> >> >> >> > Can anyone tell me how to properly launch the molpro executable in
> >> >> >> > this case?  Is it possible to use the molpro script?
> >> >> >> >
> >> >> >> >
> >> >
> >> > --
> >> > Shenggang Li
> >> > Shelby Hall L103A
> >> > Department of Chemistry
> >> > University of Alabama
> >> > Tuscaloosa, AL 35487-0336
> >> > Office: 205-348-4906
> >> >
> >>
> >>
> >>
> >>
> >
> >
> > --
> > Shenggang Li
> > Shelby Hall L103A
> > Department of Chemistry
> > University of Alabama
> > Tuscaloosa, AL 35487-0336
> > Office: 205-348-4906
> >
>
>
>
>


--
Shenggang Li
Shelby Hall L103A
Department of Chemistry
University of Alabama
Tuscaloosa, AL 35487-0336
Office: 205-348-4906