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

Sockets not working? Can't be true!
by Alexander Supalov 23 May '18

23 May '18

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

Re: [mpich-discuss] AERMOD MPI Error
by Kenneth Raffenetti 22 May '18

22 May '18

Due to limited developer resources, we no longer support Windows. We recommend using Microsoft MPI: https://msdn.microsoft.com/en-us/library/bb524831(v=vs.85).aspx Ken On 05/19/2018 12:24 PM, Tolga Pilevneli wrote: > Hello, > > I am trying to use MPICH2.1.4.1.p1 with Lakes Environmental AERMOD View. > I used this feature many times with different computers and had no > problems. > > However I recently bought a new computer (Dell Inspiron 7577 i7-7700HQ > with SSD) and did a clean Win 10 installation (v1803). I have also > installed .net framework 3.5 from Settings-> Programs and Features -> > Turn Windows Features On. > > In my first try I got this error > > Unknown option -d > Error while connecting to host, No connection could be made because the > target machine actively refused it (10061) > Connect on sock (host=R2D2, port=8676) failed, exhaused all end points > Unable to connect to 'R2D2:8676', sock error: Error = -1 > Readfile() failed, error 109 > unable to start the local smpd manager > > Then I installed mpich2 from command prompt by "msiexec /I %path%". This > time I got another error: > > Unable to open the HKEY_LOCAL_MACHINE\SOFTFWARE\MPICH\SMPD\Process\2776 > registry key, error 5, Access is denied. > > This has various similar errors with different registry keys. 3240, > 6304, 11892, 8504, 12536, 9664, 3104. Numbers change with each try. > > I have also tried various different things. Adding bin folder to system > path, giving full access to smpd from registry, different win 10 > versions (education, pro - old and recent updates), disabling firewall, > running as administrator. > > When I gave reg access and run as admin, errors stopped but Aermod > didn't run. So giving full access doesn't solve my problem. > > If you have any suggestions for this problem, I would like to try. I > need to mention my computer knowledge is only end-user level so I am not > familiar with technical terms. > > Thanks, > > Tolga > > > _______________________________________________ > 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

AERMOD MPI Error
by Tolga Pilevneli 19 May '18

19 May '18

Hello, I am trying to use MPICH2.1.4.1.p1 with Lakes Environmental AERMOD View. I used this feature many times with different computers and had no problems. However I recently bought a new computer (Dell Inspiron 7577 i7-7700HQ with SSD) and did a clean Win 10 installation (v1803). I have also installed .net framework 3.5 from Settings-> Programs and Features -> Turn Windows Features On. In my first try I got this error Unknown option -d Error while connecting to host, No connection could be made because the target machine actively refused it (10061) Connect on sock (host=R2D2, port=8676) failed, exhaused all end points Unable to connect to 'R2D2:8676', sock error: Error = -1 Readfile() failed, error 109 unable to start the local smpd manager Then I installed mpich2 from command prompt by "msiexec /I %path%". This time I got another error: Unable to open the HKEY_LOCAL_MACHINE\SOFTFWARE\MPICH\SMPD\Process\2776 registry key, error 5, Access is denied. This has various similar errors with different registry keys. 3240, 6304, 11892, 8504, 12536, 9664, 3104. Numbers change with each try. I have also tried various different things. Adding bin folder to system path, giving full access to smpd from registry, different win 10 versions (education, pro - old and recent updates), disabling firewall, running as administrator. When I gave reg access and run as admin, errors stopped but Aermod didn't run. So giving full access doesn't solve my problem. If you have any suggestions for this problem, I would like to try. I need to mention my computer knowledge is only end-user level so I am not familiar with technical terms. Thanks, Tolga _______________________________________________ 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] MPICH 3.2.1 and Slurm 14.03
by Kenneth Raffenetti 18 May '18

18 May '18

Presumably this is a batch job and not interactive? As far as I know, your job will end when the job script returns. There is no interaction with mpiexec. A job script can run mpiexec multiple times within the same allocation. Ken On 05/17/2018 03:54 AM, Bertini, Denis Dr. wrote: > Hi > I tried to use MPICH 3.2.1 together with a quite old version of SLURM ( > 14.03 ) and i notice that eventhough my > MPI program finished , SLURM still shows it as RUNNING state. > According to the MPICH documentation nothing has to be added in the > configure in order to use MPICH and SLURM > scheduler together. Did i miss something here ? Should we update to new > SLURM version? > Thanks for the support, > > Denis Bertini > IT-GSI Darmstadt > > > _______________________________________________ > 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

MPICH 3.2.1 and Slurm 14.03
by Bertini, Denis Dr. 17 May '18

17 May '18

Hi I tried to use MPICH 3.2.1 together with a quite old version of SLURM ( 14.03 ) and i notice that eventhough my MPI program finished , SLURM still shows it as RUNNING state. According to the MPICH documentation nothing has to be added in the configure in order to use MPICH and SLURM scheduler together. Did i miss something here ? Should we update to new SLURM version? Thanks for the support, Denis Bertini IT-GSI Darmstadt _______________________________________________ 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] mpirun not working
by Kenneth Raffenetti 15 May '18

15 May '18

How did you install MPICH? Were you able to successfully use the mpicc and friends compiler wrappers to build your code? Ken On 05/15/2018 01:30 AM, phani sri wrote: > Hi all, I am working on FLASH hydrodynamics, my problem is I am trying > to do parallel run in FLASH but it is showing that mpirun is not found. > I have installed the mpich and the serial run is working well. Can > anyone give a suggestion please. > ThankYou > > > > > > -- > D. P. S. L. Kameswari, > Ph. D Scholar (Physics) > ACRHEM > University Of Hyderabad, > Hyderabad, > e-mail:[email protected] <mailto:id%[email protected]> > > > _______________________________________________ > 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

mpirun not working
by phani sri 15 May '18

15 May '18

Hi all, I am working on FLASH hydrodynamics, my problem is I am trying to do parallel run in FLASH but it is showing that mpirun is not found. I have installed the mpich and the serial run is working well. Can anyone give a suggestion please. ThankYou -- D. P. S. L. Kameswari, Ph. D Scholar (Physics) ACRHEM University Of Hyderabad, Hyderabad, e-mail:[email protected] <id%3Aphanisri123(a)gmail.com> _______________________________________________ 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] mpich-3.2.1 install fail (on Ubuntu 16.04)
by Balaji, Pavan 14 May '18

14 May '18

autogen.sh is only for developers and you shouldn't have to run that. This typically happens if your timestamps are messed up, e.g., you untarred the download on one machine and tried to build it on another and the timestamps between the two machines don't match. -- Pavan > On May 4, 2018, at 12:38 PM, Jan Marek Szubert <jan.szubert(a)gmail.com> wrote: > > > > make fais when : tries to open a non-existent dir on /tmp/ > > I have spent a few days trying to determine the reason for the failure. If anything should be clear from the attached log files I would appreciate the heads up. > > The config.log file is nearly half a megabyte, otherwise, I would have included it also. I am still wading through the strace output, but it is rather overwhelming. > > > > gcc (Ubuntu 5.4.0-6ubuntu1~16.04.9) 5.4.0 20160609 > > configure and make output files attached. > > > > CC src/util/other/lib_libmpi_la-assert.lo > ( cd . && ./maint/extractcvars --dirs="`cat ./maint/cvardirs`") > Error: open directory /tmp/_HMuTRgPPE/mpich-3.2.1/maint/../src/mpi -- No such file or directory > Makefile:39284: recipe for target 'src/util/cvar/mpich_cvars.c' failed > make[2]: *** [src/util/cvar/mpich_cvars.c] Error 2 > make[2]: Leaving directory '/home/jan/amber18/AmberTools/src/mpich-3.2.1' > Makefile:38608: recipe for target 'all-recursive' failed > make[1]: *** [all-recursive] Error 1 > make[1]: Leaving directory '/home/jan/amber18/AmberTools/src/mpich-3.2.1' > Makefile:10337: recipe for target 'all' failed > make: *** [all] Error 2 > > below is the make file code close to line 39284. I have underlined the corresponding log file output and the make file line 39284 > > > # normally built by autogen.sh, but this rebuild rule is here > > $(top_srcdir)/src/util/cvar/mpich_cvars.c: $(top_srcdir)/maint/extractcvars > > ( cd $(top_srcdir) && $(top_srcdir)/maint/extractcvars --dirs="`cat $(top_srcdir)/maint/cvardirs`") > > $(cxx_buildiface_out_files): src/binding/cxx/buildiface-stamp > src/binding/cxx/buildiface-stamp: > $(top_srcdir)/src/binding/cxx/buildiface $(top_srcdir)/src/include/mpi.h.in > > ( cd $(top_srcdir)/src/binding/cxx && ./buildiface -nosep -initfile=cxx.vlist ) > > If you are able to provide any pointers it would be much appreciated.. > > Jan Szubert > > > <c.txt><m.txt>_______________________________________________ > 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] datatypes and MPI-3 shared memory windows
by Min Si 10 May '18

10 May '18

Hi Lux, You can still use MPI send/receive or MPI RMA with datatypes to exchange the halo region in a shared window, similar to the flat-MPI version. But I think you might want to completely avoid extra copies for intra-node neighbors when using shared window. I.e., instead of copying the edge elements to neighbors' halo, a process can directly read the edge elements owned by a neighbor if it is in the same shared window. Thus, you use MPI communication with datatypes only for neighbors who are on remote node. Min On 2018/05/10 12:00, Luis Cebamanos wrote: > Hi Min, > > Thanks for your help. My intention is to use shared mem windows in a 3D > halo exchange communication. The application already uses complex MPI > datatypes for the exchange of edges and corners. I guess my question is > how should I deal with this? Should I manually unpack the MPI datatypes > into a shared memory buffer, then communicate (intra- and inter- node) > and finally pack the data back to its original data structure so the > simulation can carry on with other operations. > > double *baseptr; > MPI_Win_allocate_shared (len, 1, MPI_INFO_NULL, mpi3->shmcomm, > &baseptr, &mpi3->win); > > Pointer baseptr gives me the address to the shared mem window, but I > cannot see the way of using datatypes in the shared mem created. > > Thanks > Lux > > On 10/05/2018 16:53, Min Si wrote: >> Hi Lux, >> >> Yes, you can use MPI datatype with shared window. Can you provide more >> information about the purpose to use shared window thus we can give >> appropriate suggestion ? >> >> Best regards, >> Min >> >> On 2018/05/10 10:05, Luis Cebamanos wrote: >>> Hi all, >>> >>> Is there a way to deal with MPI datatypes when using shmm windows? What >>> is the advice to convert a flat MPI code that uses MPI datatypes for >>> halo exchanges to use shmm windows? >>> >>> Thanks! >>> Lux >>> >>> >>> _______________________________________________ >>> 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 _______________________________________________ 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] datatypes and MPI-3 shared memory windows
by Luis Cebamanos 10 May '18

10 May '18

Hi Min, Thanks for your help. My intention is to use shared mem windows in a 3D halo exchange communication. The application already uses complex MPI datatypes for the exchange of edges and corners. I guess my question is how should I deal with this? Should I manually unpack the MPI datatypes into a shared memory buffer, then communicate (intra- and inter- node) and finally pack the data back to its original data structure so the simulation can carry on with other operations. double *baseptr; MPI_Win_allocate_shared (len, 1, MPI_INFO_NULL, mpi3->shmcomm, &baseptr, &mpi3->win); Pointer baseptr gives me the address to the shared mem window, but I cannot see the way of using datatypes in the shared mem created. Thanks Lux On 10/05/2018 16:53, Min Si wrote: > Hi Lux, > > Yes, you can use MPI datatype with shared window. Can you provide more > information about the purpose to use shared window thus we can give > appropriate suggestion ? > > Best regards, > Min > > On 2018/05/10 10:05, Luis Cebamanos wrote: >> Hi all, >> >> Is there a way to deal with MPI datatypes when using shmm windows? What >> is the advice to convert a flat MPI code that uses MPI datatypes for >> halo exchanges to use shmm windows? >> >> Thanks! >> Lux >> >> >> _______________________________________________ >> 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