Install mpiexe on SGI Onyx

[William Wei]

Thank you, guys. With GNU make, I went through the compiling.  Now I have
difficulty to run my program:

Qsub_Script:
#!/bin/csh

#PBS -V
#PBS -r n
#PBS -N Anti_Dyn
#PBS -l cput=6000:00:00
#PBS -l ncpus=4,walltime=720:00:00
#PBS -m abe
#PBS -M mdit at phm.utoronto.ca
#PBS -q long
#

set WRKDIR=/aspirin/william/Interferon/Model_Jyothi/MDyn_FitIFN

/aspirin/software/MPIexe/mpiexec-0.75/mpiexec mpirun -np 4 \
        $AMBERHOME/exe/sander -O -i $WRKDIR/Anti_IFN_CnsV_10ps.in \
                              -o $WRKDIR/Anti_IFN_CnsV_10ps.out \
                              -p $WRKDIR/Anti.top \
                              -c $WRKDIR/Anti_Min_20K_rst \
                              -x $WRKDIR/Anti_IFN_ConV_10ps_mdcrd \
                              -v $WRKDIR/Anti_IFN_ConV_10ps_mdvel \
                              -e $WRKDIR/Anti_IFN_ConV_10ps_mden \
                              -r $WRKDIR/Anti_IFN_ConV_10ps_rst \
                              -inf $WRKDIR/Anti_IFN_ConV_10ps_mdinfo


Ourput:
Word too long.
/bin/sh: Arg list too long
/bin/sh: Arg list too long
mpiexec: Warning: reconnect_to_mom: task 0 not found, assuming done
.
mpiexec: Warning: reconnect_to_mom: task 0 not found, assuming done
.
mpiexec: Error: start_tasks: tm_obit task 2: tm: not found.

Could you please give me some idea? Thank you all.

Have a good day,
William

-----Original Message-----
From: Troy Baer [mailto:troy at osc.edu]
Sent: Monday, February 16, 2004 4:35 PM
To: William Wei
Subject: RE: Install mpiexe on SGI Onyx


On Mon, 2004-02-16 at 16:16, William Wei wrote:
> Thank you for the information. With your suggestion I went little further,
> but still the make tell me:
>
>
gcc -MM -g -O2 -Wall -W -Wpointer-arith -Wwrite-strings -Wcast-align -Wcast-
>
qual -Wbad-function-cast -Wundef -Wmissing-prototypes -Wmissing-declarations
>  -Wnested-externs -I/usr/pbs/include mpiexec.c get_hosts.c start_tasks.c
> wait_tasks.c util.c config.c stdio.c growstr.c > .depend
>
gcc -g -O2 -Wall -W -Wpointer-arith -Wwrite-strings -Wcast-align -Wcast-qual
>  -Wbad-function-cast -Wundef -Wmissing-prototypes -Wmissing-declarations -
Wn
> ested-externs -I/usr/pbs/include -c mpiexec.c
>
gcc -g -O2 -Wall -W -Wpointer-arith -Wwrite-strings -Wcast-align -Wcast-qual
>  -Wbad-function-cast -Wundef -Wmissing-prototypes -Wmissing-declarations -
Wn
> ested-externs -I/usr/pbs/include -c get_hosts.c
>
gcc -g -O2 -Wall -W -Wpointer-arith -Wwrite-strings -Wcast-align -Wcast-qual
>  -Wbad-function-cast -Wundef -Wmissing-prototypes -Wmissing-declarations -
Wn
> ested-externs -I/usr/pbs/include -c start_tasks.c
> start_tasks.c: In function `scatter_gm_startup_ports':
> start_tasks.c:959: variable `tv' has initializer but incomplete type
> start_tasks.c:959: warning: excess elements in struct initializer
> start_tasks.c:959: warning: (near initialization for `tv')
> start_tasks.c:959: warning: excess elements in struct initializer
> start_tasks.c:959: warning: (near initialization for `tv')
> start_tasks.c:959: storage size of `tv' isn't known
> start_tasks.c:959: warning: unused variable `tv'
> make: *** [start_tasks.o] Error 1
>
> It become more complicated.

It looks like the compiler is missing the definition of what a "struct
timeval" is.  Add the following line to the set of #include's at the
beginning of start_tasks.c:

#include <sys/time.h>

I think that should fix it, although I don't have an Irix box with PBS
handy to test it.  I'll let the developer know about this too.

> My original problem is: within PBS system (I am using PBSPRO5.3.3), after
I
> removed the jobs using qdel, the job was removed from PBS queue. But it
> still was running under shell (I could see it from 'top').  Did you meet
> this problem before? Thanks.

That problem is why we developed mpiexec in the first place.

	--Troy
--
Troy Baer                       email:  troy at osc.edu
Science & Technology Support    phone:  614-292-9701
Ohio Supercomputer Center       web:  http://oscinfo.osc.edu