June 2018 - discuss - mailman3.cels.anl.gov

Re: [mpich-discuss] Waitany for Ibarrier and Irecv in sequential program
by Nils-Arne Dreier 07 Jun '18

07 Jun '18

I just figured it out my self: MPI_Ibarrier returns MPI_REQUEST_NULL. Sorry for the noise! Best, Nils On 07.06.2018 11:51, Nils-Arne Dreier wrote: > Dear MPICH community, > > the following program deadlocks when executed with MPICH 3.2 (Debian > stretch) on a single rank. > > > barrier_waitany.cc > #include <mpi.h> > > int main(int argc, char** argv){ > MPI_Request req[2]; > int buf; > int index; > MPI_Init(&argc, &argv); > MPI_Ibarrier(MPI_COMM_WORLD, &req[0]); > MPI_Irecv(&buf, 1, MPI_INT, 0, 0, MPI_COMM_WORLD, &req[1]); > MPI_Waitany(2, req, &index, MPI_STATUS_IGNORE); > MPI_Finalize(); > return 0; > } > > Compile with: > mpic++ -g -std=c++14 -Wall mpi_barrier_waitany.cc -o mpi_barrier_waitany > > Execute with: > mpirun -n 1 ./barrier_waitany > > Best, > Nils > > > > > _______________________________________________ > 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

Waitany for Ibarrier and Irecv in sequential program
by Nils-Arne Dreier 07 Jun '18

07 Jun '18

Dear MPICH community, the following program deadlocks when executed with MPICH 3.2 (Debian stretch) on a single rank. barrier_waitany.cc #include <mpi.h> int main(int argc, char** argv){ MPI_Request req[2]; int buf; int index; MPI_Init(&argc, &argv); MPI_Ibarrier(MPI_COMM_WORLD, &req[0]); MPI_Irecv(&buf, 1, MPI_INT, 0, 0, MPI_COMM_WORLD, &req[1]); MPI_Waitany(2, req, &index, MPI_STATUS_IGNORE); MPI_Finalize(); return 0; } Compile with: mpic++ -g -std=c++14 -Wall mpi_barrier_waitany.cc -o mpi_barrier_waitany Execute with: mpirun -n 1 ./barrier_waitany Best, Nils _______________________________________________ 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] Sockets not working? Can't be true!
by Alexander Supalov 06 Jun '18

06 Jun '18

Thanks. Will report back once done. 5 июня 2018 г., в 17:36, Kenneth Raffenetti <raffenet(a)mcs.anl.gov> написал(а): Hi Alexander, A fix for this was included in version 3.2.1. Let us know if you still see an issue after updating. Ken On 05/23/2018 06:32 AM, Alexander Supalov wrote: Hi everybody, I made a funny observation using a simple ping-pong MPI program on my local network setup. Intranode runs are OK. Internode runs hang over ch3:sock and fail or hang over nemesis, depending on OS version. Standard ping works fine internode on both sides, and so do my own socket-based programs. Any idea what this might be? Nemesis diagnostics under Ubuntu 16.04 LTS: c -f mpi.hosts -n 2 $PWD/pingpong Assertion failed in file src/mpid/ch3/channels/nemesis/netmod/tcp/socksm.c at line 597: hdr.datalen != 0 internal ABORT - process 1 Host list: 192.168.178.31 192.168.178.32 Platform: Intel, Ubuntu 16.04 LTS on one side, Ubuntu 14.04 LTS on the other, MPICH-3.2 on both, 100-Mbit Ethernet in between. Best regards. Alexander _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

couldn't built mpich-master on SuSE Linux with Sun C
by Siegmar Gross 06 Jun '18

06 Jun '18

Hi, I tried to build mpich-master on my "SUSE Linux Enterprise Server 12.3 (x86_64)" with Sun C 5.15 (Oracle Developer Studio 12.6). Unfortunately, I get the following error, that I already reported already on May 9th. loki mpich-master_2018-06-06_Linux.x86_64.64_cc 122 tail -22 log.make.Linux.x86_64.64_cc | grep -v NOTICE "../mpich/src/include/mpir_handlemem.h", line 61: warning: null dimension: indirect "../mpich/src/include/mpir_handlemem.h", line 122: warning: null dimension: indirect MOD src/binding/fortran/use_mpi/mpi.mod-stamp "./src/include/mpichconf.h", line 1708: Warning (Anachronism): Attempt to redefine __restrict__ without using #undef. 1 Warning(s) detected. f90: Warning: Option -shared passed to ld, if ld is invoked, ignored otherwise f90: Warning: Option --whole-archive passed to ld, if ld is invoked, ignored otherwise f90: Warning: Option --no-whole-archive passed to ld, if ld is invoked, ignored otherwise f90: Warning: Option -path passed to ld, if ld is invoked, ignored otherwise f90: Warning: Option -path passed to ld, if ld is invoked, ignored otherwise f90: Warning: Option -soname passed to ld, if ld is invoked, ignored otherwise /usr/bin/ld: -r and -shared may not be used together postopt: error: ld failed to link the binary Makefile:16787: recipe for target 'lib/libmpi.la' failed make[2]: *** [lib/libmpi.la] Error 2 make[2]: Leaving directory '/export2/src/mpich-master/mpich-master_2018-06-06_Linux.x86_64.64_cc' Makefile:39797: recipe for target 'all-recursive' failed make[1]: *** [all-recursive] Error 1 make[1]: Leaving directory '/export2/src/mpich-master/mpich-master_2018-06-06_Linux.x86_64.64_cc' Makefile:10453: recipe for target 'all' failed make: *** [all] Error 2 loki mpich-master_2018-06-06_Linux.x86_64.64_cc 123 I used the following commands to update and configure MPICH. cd mpich git pull https://github.com/pmodels/mpich.git master git submodule update --init --recursive autogen.sh mkdir mpich-master_`date -I'date'`_${SYSTEM_ENV}.${MACHINE_ENV}.64_cc cd mpich-master_`date -I'date'`_${SYSTEM_ENV}.${MACHINE_ENV}.64_cc ../mpich/configure \ --prefix=/usr/local/mpich-master_64_cc \ --libdir=/usr/local/mpich-master_64_cc/lib64 \ --includedir=/usr/local/mpich-master_64_cc/include64 \ CC="cc" CXX="CC" F77="f77" FC="f95" \ CFLAGS="-m64" CXXFLAGS="-m64" FFLAGS="-m64" FCFLAGS="-m64" \ LDFLAGS="-m64 -L/usr/lib64 -R/usr/lib64" \ --enable-fortran=yes --enable-cxx --enable-romio \ --enable-debuginfo --enable-smpcoll \ --enable-threads=multiple --with-thread-package=posix \ --enable-shared \ |& tee log.configure.$SYSTEM_ENV.$MACHINE_ENV.64_cc make -j 6 |& tee log.make.$SYSTEM_ENV.$MACHINE_ENV.64_cc I would be grateful, if somebody can fix the problem or is it a problem of the Oracle compiler?. Do you need anything else? Thank you very much for any help in advance. Kind regards Siegmar _______________________________________________ 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] Role of ssh tunnel for hydra
by Kenneth Raffenetti 05 Jun '18

05 Jun '18

Hi Abu, No, internal hydra communication is not tunneled. If you launch your application with 'mpiexec -v', you can see that a "control port" is supplied to each proxy, which is then used to connect back to mpiexec. This TCP port is from the unprivileged range. Ken On 06/05/2018 09:38 AM, Abu Naser wrote: > Good Morning, > > > I am trying to understand the role of job launcher (e.g. ssh ) in MPICH. > I have looked over the mpich design documents and hydra process > management doc. But did not get the answer I am looking for. Would you > please give the answer of my following question? > > > 1. If hydra uses ssh as a job launcher for multiple machines, then are > all communications done by hydra performed using ssh tunnel? In other > words, if I configure ssh in port 22, then will all communication done > by hydra with other machines use port 22? > > > > > Thank you. > > > > > Best Regards, > > Abu Naser > > > > _______________________________________________ > 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] Sockets not working? Can't be true!
by Kenneth Raffenetti 05 Jun '18

05 Jun '18

Hi Alexander, A fix for this was included in version 3.2.1. Let us know if you still see an issue after updating. Ken On 05/23/2018 06:32 AM, Alexander Supalov wrote: > Hi everybody, > > I made a funny observation using a simple ping-pong MPI program on my > local network setup. Intranode runs are OK. Internode runs hang over > ch3:sock and fail or hang over nemesis, depending on OS version. > Standard ping works fine internode on both sides, and so do my own > socket-based programs. Any idea what this might be? > > Nemesis diagnostics under Ubuntu 16.04 LTS: > > c -f mpi.hosts -n 2 $PWD/pingpong > Assertion failed in file > src/mpid/ch3/channels/nemesis/netmod/tcp/socksm.c at line 597: > hdr.datalen != 0 > internal ABORT - process 1 > > Host list: > > 192.168.178.31 > 192.168.178.32 > > Platform: > > Intel, Ubuntu 16.04 LTS on one side, Ubuntu 14.04 LTS on the other, > MPICH-3.2 on both, 100-Mbit Ethernet in between. > > Best regards. > > Alexander > > > > _______________________________________________ > 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

Role of ssh tunnel for hydra
by Abu Naser 05 Jun '18

05 Jun '18

Good Morning, I am trying to understand the role of job launcher (e.g. ssh ) in MPICH. I have looked over the mpich design documents and hydra process management doc. But did not get the answer I am looking for. Would you please give the answer of my following question? 1. If hydra uses ssh as a job launcher for multiple machines, then are all communications done by hydra performed using ssh tunnel? In other words, if I configure ssh in port 22, then will all communication done by hydra with other machines use port 22? Thank you. Best Regards, Abu Naser _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0