April 2014 - discuss - mailman3.cels.anl.gov

Re: [mpich-discuss] file open error code checking
by Wei-keng Liao 07 Apr '14

07 Apr '14

In my opinion, it is best to do this checking in common code, e.g. common/ad_open etc. For those file systems that their ADIO drivers make a call to open() and set errno correctly (following POSIX convention), this is straightforward. But for those file systems, such as PVFS2, that do not make calls to open(), their drivers must take care of the errno separately. Fortunately, those file systems are rare. I can see this checking currently is all over places. For easy software management, it will be great if ROMIO keeps it in ad_open only (assuming all ADIO drivers first call ad_open before their platform-specific ad_xxx_open). Wei-keng On Apr 7, 2014, at 9:16 AM, Rob Latham wrote: > > > On 04/06/2014 12:23 PM, Wei-keng Liao wrote: >> I can see error codes EISDIR and EEXIST are checked in ad_ufs_open.c >> and the corresponding MPI_ERR_BAD_FILE and MPI_ERR_FILE_EXISTS are generated. >> >> But this update does not appear on other file systems, such as Lustre and GPFS. >> PnetCDF needs an MPI-IO that can return the MPI_ERR_FILE_EXISTS class when the >> MPI_MODE_EXCL mode is set and the file exist. This class will correspond to >> NC_EEXIST in PnetCDF (when NC_NOCLOBBER is set in ncmpi_create and file exists). >> > > Thanks, Wei-keng. I wonder if we can do this error code setting up in common code so all the lower-level file system drivers get it automatically? > > ==rob > > >> >> ad_ufs_open.c >> >> 94 else if(errno == EISDIR) { >> 95 *error_code = MPIO_Err_create_code(MPI_SUCCESS, >> 96 MPIR_ERR_RECOVERABLE, myname, >> 97 __LINE__, MPI_ERR_BAD_FILE, >> 98 "**filename", 0); >> 99 } >> 100 else if(errno == EEXIST) { >> 101 *error_code = MPIO_Err_create_code(MPI_SUCCESS, >> 102 MPIR_ERR_RECOVERABLE, myname, >> 103 __LINE__, MPI_ERR_FILE_EXISTS, >> 104 "**fileexist", 0); >> 105 >> 106 } >> >> >> Wei-keng >> >> _______________________________________________ >> discuss mailing list discuss(a)mpich.org >> To manage subscription options or unsubscribe: >> https://lists.mpich.org/mailman/listinfo/discuss >> > > -- > Rob Latham > Mathematics and Computer Science Division > Argonne National Lab, IL USA > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] statically linked mpiexec
by Kenneth Raffenetti 07 Apr '14

07 Apr '14

Ah, sorry. I misread your first email and assumed you were only talking about the MPICH libraries and compiler wrappers. The Intel compilers should accept a '-static' flag that will force link all libraries statically. Try adding it to your CFLAGS to force this behavior on mpiexec. Ken On 04/07/2014 09:50 AM, Andy May wrote: > Ken, > > The '--disable-shared' flag removes some of the external references (to > the Intel compiler in my case) but still leaves: > > ldd ./bin/mpiexec > linux-vdso.so.1 (0x00007fff7a7b9000) > libnsl.so.1 => /lib64/libnsl.so.1 (0x00007f3298e3f000) > libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f3298c21000) > libpciaccess.so.0 => /usr/lib64/libpciaccess.so.0 > (0x00007f3298a17000) > libxml2.so.2 => /usr/lib64/libxml2.so.2 (0x00007f32986ac000) > libdl.so.2 => /lib64/libdl.so.2 (0x00007f32984a8000) > libz.so.1 => /lib64/libz.so.1 (0x00007f3298292000) > liblzma.so.5 => /usr/lib64/liblzma.so.5 (0x00007f329806c000) > libm.so.6 => /lib64/libm.so.6 (0x00007f3297d69000) > libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x00007f3297b52000) > libc.so.6 => /lib64/libc.so.6 (0x00007f32977a3000) > /lib64/ld-linux-x86-64.so.2 (0x00007f3299057000) > > whereas what I did gives: > > ldd ./bin/mpiexec > not a dynamic executable > > Is there any option to ensure fully statically linked mpiexec via > configure? > > Thanks, > > Andy > > On 07/04/14 13:26, Kenneth Raffenetti wrote: >> Andy, >> >> The official way is to add '--disable-shared' to your configure line. >> >> Ken >> >> On 04/07/2014 05:47 AM, Andy May wrote: >>> Hi, >>> >>> Does anyone know if there is a supported or recommended way to build >>> MPICH such that it's binaries (eg. mpiexec) are statically linked? >>> >>> At present, post building I'm running: >>> >>> cd src/pm/hydra >>> eval ./configure `./config.status --config` --disable-pci >>> --disable-libxml2 >>> make CCLD='$(CC) -all-static' >>> >>> which just disables two cases where the static libraries aren't >>> available, then links statically, but is there a better or correct way >>> to do this? >>> >>> Many thanks, >>> >>> Andy >>> _______________________________________________ >>> discuss mailing list discuss(a)mpich.org >>> To manage subscription options or unsubscribe: >>> https://lists.mpich.org/mailman/listinfo/discuss >> _______________________________________________ >> discuss mailing list discuss(a)mpich.org >> To manage subscription options or unsubscribe: >> https://lists.mpich.org/mailman/listinfo/discuss > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] statically linked mpiexec
by Andy May 07 Apr '14

07 Apr '14

Ken, The '--disable-shared' flag removes some of the external references (to the Intel compiler in my case) but still leaves: ldd ./bin/mpiexec linux-vdso.so.1 (0x00007fff7a7b9000) libnsl.so.1 => /lib64/libnsl.so.1 (0x00007f3298e3f000) libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f3298c21000) libpciaccess.so.0 => /usr/lib64/libpciaccess.so.0 (0x00007f3298a17000) libxml2.so.2 => /usr/lib64/libxml2.so.2 (0x00007f32986ac000) libdl.so.2 => /lib64/libdl.so.2 (0x00007f32984a8000) libz.so.1 => /lib64/libz.so.1 (0x00007f3298292000) liblzma.so.5 => /usr/lib64/liblzma.so.5 (0x00007f329806c000) libm.so.6 => /lib64/libm.so.6 (0x00007f3297d69000) libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x00007f3297b52000) libc.so.6 => /lib64/libc.so.6 (0x00007f32977a3000) /lib64/ld-linux-x86-64.so.2 (0x00007f3299057000) whereas what I did gives: ldd ./bin/mpiexec not a dynamic executable Is there any option to ensure fully statically linked mpiexec via configure? Thanks, Andy On 07/04/14 13:26, Kenneth Raffenetti wrote: > Andy, > > The official way is to add '--disable-shared' to your configure line. > > Ken > > On 04/07/2014 05:47 AM, Andy May wrote: >> Hi, >> >> Does anyone know if there is a supported or recommended way to build >> MPICH such that it's binaries (eg. mpiexec) are statically linked? >> >> At present, post building I'm running: >> >> cd src/pm/hydra >> eval ./configure `./config.status --config` --disable-pci >> --disable-libxml2 >> make CCLD='$(CC) -all-static' >> >> which just disables two cases where the static libraries aren't >> available, then links statically, but is there a better or correct way >> to do this? >> >> Many thanks, >> >> Andy >> _______________________________________________ >> discuss mailing list discuss(a)mpich.org >> To manage subscription options or unsubscribe: >> https://lists.mpich.org/mailman/listinfo/discuss > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] file open error code checking
by Rob Latham 07 Apr '14

07 Apr '14

On 04/06/2014 12:23 PM, Wei-keng Liao wrote: > I can see error codes EISDIR and EEXIST are checked in ad_ufs_open.c > and the corresponding MPI_ERR_BAD_FILE and MPI_ERR_FILE_EXISTS are generated. > > But this update does not appear on other file systems, such as Lustre and GPFS. > PnetCDF needs an MPI-IO that can return the MPI_ERR_FILE_EXISTS class when the > MPI_MODE_EXCL mode is set and the file exist. This class will correspond to > NC_EEXIST in PnetCDF (when NC_NOCLOBBER is set in ncmpi_create and file exists). > Thanks, Wei-keng. I wonder if we can do this error code setting up in common code so all the lower-level file system drivers get it automatically? ==rob > > ad_ufs_open.c > > 94 else if(errno == EISDIR) { > 95 *error_code = MPIO_Err_create_code(MPI_SUCCESS, > 96 MPIR_ERR_RECOVERABLE, myname, > 97 __LINE__, MPI_ERR_BAD_FILE, > 98 "**filename", 0); > 99 } > 100 else if(errno == EEXIST) { > 101 *error_code = MPIO_Err_create_code(MPI_SUCCESS, > 102 MPIR_ERR_RECOVERABLE, myname, > 103 __LINE__, MPI_ERR_FILE_EXISTS, > 104 "**fileexist", 0); > 105 > 106 } > > > Wei-keng > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss > -- Rob Latham Mathematics and Computer Science Division Argonne National Lab, IL USA

1 0

Re: [mpich-discuss] Multiple passphrase queries
by "Antonio J. Peña" 07 Apr '14

07 Apr '14

Sorry, we stopped supporting Windows directly a some time ago. Please refer to Microsoft MPI, based on MPICH. You have a link in our downloads page http://www.mpich.org/downloads/ Antonio On 04/07/2014 08:34 AM, Alexander Pristavko wrote: > I've installed mpich2 on my windows 8.1 machine, saved my account > login and password to register. Everything went successfully until I > decided to run the cpi.exe example. When I try to launch it in > 1-processor mode it asks passphrase and launches application after > I've type in the passphrase. > But when I try to launch the application in n-processor mode it asks > passphrase n times at once. So it looks like on the attached > screenshot. And nothing works from here. Even typing the passphrase n > times. > Please help me resolve this problem. > Thank you. > > -- > Sincerely, chestercore > > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss -- Antonio J. Peï¿½a Postdoctoral Appointee Mathematics and Computer Science Division Argonne National Laboratory 9700 South Cass Avenue, Bldg. 240, Of. 3148 Argonne, IL 60439-4847 apenya(a)mcs.anl.gov www.mcs.anl.gov/~apenya

1 0

Multiple passphrase queries
by Alexander Pristavko 07 Apr '14

07 Apr '14

I've installed mpich2 on my windows 8.1 machine, saved my account login and password to register. Everything went successfully until I decided to run the cpi.exe example. When I try to launch it in 1-processor mode it asks passphrase and launches application after I've type in the passphrase. But when I try to launch the application in n-processor mode it asks passphrase n times at once. So it looks like on the attached screenshot. And nothing works from here. Even typing the passphrase n times. Please help me resolve this problem. Thank you. -- Sincerely, chestercore

1 0

Re: [mpich-discuss] statically linked mpiexec
by Kenneth Raffenetti 07 Apr '14

07 Apr '14

Andy, The official way is to add '--disable-shared' to your configure line. Ken On 04/07/2014 05:47 AM, Andy May wrote: > Hi, > > Does anyone know if there is a supported or recommended way to build > MPICH such that it's binaries (eg. mpiexec) are statically linked? > > At present, post building I'm running: > > cd src/pm/hydra > eval ./configure `./config.status --config` --disable-pci --disable-libxml2 > make CCLD='$(CC) -all-static' > > which just disables two cases where the static libraries aren't > available, then links statically, but is there a better or correct way > to do this? > > Many thanks, > > Andy > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] statically linked mpiexec
by Wesley Bland 07 Apr '14

07 Apr '14

When you run the configure script, use --enable-static and that should take care of what you're looking for. On Monday, April 7, 2014, Andy May <MayAJ1(a)cardiff.ac.uk> wrote: > Hi, > > Does anyone know if there is a supported or recommended way to build MPICH > such that it's binaries (eg. mpiexec) are statically linked? > > At present, post building I'm running: > > cd src/pm/hydra > eval ./configure `./config.status --config` --disable-pci --disable-libxml2 > make CCLD='$(CC) -all-static' > > which just disables two cases where the static libraries aren't available, > then links statically, but is there a better or correct way to do this? > > Many thanks, > > Andy > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss >

1 0

Re: [mpich-discuss] mpich2 - checkpointing error
by Wesley Bland 07 Apr '14

07 Apr '14

Unfortunately, this is a known problem at the moment. BLCR checkpointing hasn't worked for a few versions now. It's something we're working to fix in a future version. Thanks, Wesley On Monday, April 7, 2014, Marcelo Paiva Ramos <marcelo.paiva(a)cptec.inpe.br> wrote: > Hi, > Can you help me to solve this problem? > > cat /etc/issue > CentOS release 6.5 (Final) > Kernel \r on an \m > > uname -a > Linux server 2.6.32-431.11.2.el6.x86_64 #1 SMP Tue Mar 25 19:59:55 UTC > 2014 x86_64 x86_64 x86_64 GNU/Linux > > *INSTALL: blcr-0.8.5* > tar xzvf blcr-0.8.5.tar.gz > cd blcr-0.8.5 > mkdir builddir > cd builddir > ../configure --prefix=/opt/blcr > make > make install > /sbin/insmod /opt/blcr/lib/blcr/2.6.32-431.11.2.el6.x86_64/blcr_imports.ko > /sbin/insmod /opt/blcr/lib/blcr/2.6.32-431.11.2.el6.x86_64/blcr.ko > uname -r > 2.6.32-431.11.2.el6.x86_64 > lsmod | grep blcr > blcr 115465 0 > blcr_imports 10715 1 blcr > ldconfig -p | grep blcr > libcr_run.so.0 (libc6,x86-64) => /opt/blcr/lib/libcr_run.so.0 > libcr_run.so (libc6,x86-64) => /opt/blcr/lib/libcr_run.so > libcr_omit.so.0 (libc6,x86-64) => /opt/blcr/lib/libcr_omit.so.0 > libcr_omit.so (libc6,x86-64) => /opt/blcr/lib/libcr_omit.so > libcr.so.0 (libc6,x86-64) => /opt/blcr/lib/libcr.so.0 > libcr.so (libc6,x86-64) => /opt/blcr/lib/libcr.so > chkconfig --list | grep blcr > blcr 0:off 1:off 2:on 3:on 4:on 5:on 6:off > > *INSTALL: mpich-3.1* > tar xzvf mpich-3.1.tar.gz > cd mpich-3.1 > ./configure --disable-fast CFLAGS=-O2 FFLAGS=-O2 CXXFLAGS=-O2 FCFLAGS=-O2 > --prefix=/opt/mpich2/ CC=/opt/intel/bin/icc FC=/opt/intel/bin/ifort > F77=/opt/intel/bin/ifort --enable-checkpointing > --with-hydra-ckpointlib=blcr --with-blcr=/opt/blcr > --with-blcr-include=/opt/blcr/include --with-blcr-lib=/opt/blcr/lib > make > make install > > *.bashrc* > export PATH=$PATH:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/ > bin:/sbin:/bin:/opt/blcr/bin:/opt/mpich2/bin > export LD_LIBRARY_PATH=/usr/local/lib:/usr/local/lib64:/opt/ > intel/lib/intel64:/opt/blcr/lib:/opt/mpich2/lib > > *mpiexec -info* > HYDRA build details: > Version: 3.1 > Release Date: Thu Feb 20 11:41:13 CST 2014 > CC: /opt/intel/bin/icc -O2 > CXX: g++ -O2 > F77: /opt/intel/bin/ifort -O2 > F90: /opt/intel/bin/ifort -O2 > Configure options: '--disable-option-checking' > '--prefix=/opt/mpich2' '--disable-fast' 'CFLAGS=-O2 -O0' 'FFLAGS=-O2 -O0' > 'CXXFLAGS=-O2 ' 'FCFLAGS=-O2 ' 'CC=/opt/intel/bin/icc' > 'FC=/opt/intel/bin/ifort' 'F77=/opt/intel/bin/ifort' > '--enable-checkpointing' '--with-hydra-ckpointlib=blcr' > '--with-blcr=/opt/blcr' '--with-blcr-include=/opt/blcr/include' > '--with-blcr-lib=/opt/blcr/lib' '--cache-file=/dev/null' '--srcdir=.' > 'LDFLAGS= -L/opt/blcr/lib' 'LIBS=-lrt -lcr -lpthread ' 'CPPFLAGS= > -I/root/mpich-3.1/src/mpl/include -I/root/mpich-3.1/src/mpl/include > -I/root/mpich-3.1/src/openpa/src -I/root/mpich-3.1/src/openpa/src > -I/root/mpich-3.1/src/mpi/romio/include -I/opt/blcr/include' > Process Manager: pmi > Launchers available: ssh rsh fork slurm ll lsf sge > pbs manual persist > Topology libraries available: hwloc > Resource management kernels available: user slurm ll lsf sge pbs > cobalt > Checkpointing libraries available: blcr > Demux engines available: poll select > > > *ERROR* > mpiexec -n 1 -ckpointlib blcr -ckpoint-interval 20 -ckpoint-prefix > /home/marcelo/TESTE/ ./teste > [proxy:0:0@server] requesting checkpoint > [proxy:0:0@server] checkpoint completed > [proxy:0:0@server] HYDT_ckpoint_blcr_checkpoint > (tools/ckpoint/blcr/ckpoint_blcr.c:241): Checkpointing failed. Make sure > BLCR kernel module is loaded. Unknown error 2356 > [proxy:0:0@server] ckpoint_thread (tools/ckpoint/ckpoint.c:76): blcr > checkpoint returned error > [proxy:0:0@server] requesting checkpoint > [proxy:0:0@server] checkpoint completed > [proxy:0:0@server] HYDT_ckpoint_blcr_checkpoint > (tools/ckpoint/blcr/ckpoint_blcr.c:241): Checkpointing failed. Make sure > BLCR kernel module is loaded. Unknown error 2356 > [proxy:0:0@server] ckpoint_thread (tools/ckpoint/ckpoint.c:76): blcr > checkpoint returned error > > Best regards, > Marcelo. >

1 0

mpiexec
by Mojtaba S. Moghani 07 Apr '14

07 Apr '14

After this commad : mpiexec -n 5 -f machinefile ./examples/cpi, it has some errors when run, errors like : HYDU_parse_hostfile (/home/mojtaba/libraries/mpich-3.1/src/pm/hydra/utils/args/args.c:319): unable to open host file: machinefile [mpiexec@Mojtaba] mfile_fn (/home/mojtaba/libraries/mpich-3.1/src/pm/hydra/ui/mpich/utils.c:336): error parsing hostfile [mpiexec@Mojtaba] match_arg (/home/mojtaba/libraries/mpich-3.1/src/pm/hydra/utils/args/args.c:152): match handler returned error [mpiexec@Mojtaba] HYDU_parse_array (/home/mojtaba/libraries/mpich-3.1/src/pm/hydra/utils/args/args.c:174): argument matching returned error [mpiexec@Mojtaba] parse_args (/home/mojtaba/libraries/mpich-3.1/src/pm/hydra/ui/mpich/utils.c:1596): error parsing input array [mpiexec@Mojtaba] HYD_uii_mpx_get_parameters (/home/mojtaba/libraries/mpich-3.1/src/pm/hydra/ui/mpich/utils.c:1648): unable to parse user arguments [mpiexec@Mojtaba] main (/home/mojtaba/libraries/mpich-3.1/src/pm/hydra/ui/mpich/mpiexec.c:153): error parsing parameters what is the problem? thanks, -- Mojtaba S. Moghani Faculty of Environment University of Tehran

1 0