June 2013 - discuss - mailman3.cels.anl.gov

Re: [mpich-discuss] Problem reproducing an example from the MPI standard
by Pavan Balaji 22 Jun '13

22 Jun '13

The example is correct, since it's in Fortran. Your conversion to C is incorrect. You probably want to use MPI_DOUBLE instead of MPI_REAL, and MPI_INT instead of MPI_INTEGER. -- Pavan On 06/22/2013 11:49 AM, Jiri Simsa wrote: > Hi, > > I tried implementing the Example 3.17 from the MPI 3.0 specification > document using as follows: > > #include <assert.h> > #include <mpi.h> > #include <stdio.h> > > int main(int argc, char *argv[]) { > int myrank, size; > MPI_Status status; > MPI_Init(&argc, &argv ); > MPI_Comm_rank(MPI_COMM_WORLD, &myrank); > MPI_Comm_size(MPI_COMM_WORLD, &size); > assert(size == 3); > if (myrank == 0) { > /* code for process zero */ > int i = 1; > MPI_Send(&i, 1, MPI_INTEGER, 2, 99, MPI_COMM_WORLD); > } > if (myrank == 1) { > /* code for process one */ > double d = 3.14; > MPI_Send(&d, 1, MPI_REAL, 2, 99, MPI_COMM_WORLD); > } > if (myrank == 2) { > /* code for process two */ > for (int i = 0; i < 2; i++) { > MPI_Probe(MPI_ANY_SOURCE, MPI_ANY_TAG, MPI_COMM_WORLD, &status); > printf("Probe matched %lld bytes from source %d.\n", > status.count, status.MPI_SOURCE); > if (status.MPI_SOURCE == 0) { > int i; > MPI_Recv(&i, 1, MPI_INTEGER, 0, 99, MPI_COMM_WORLD, &status); > printf("Received integer '%d' from %d.\n", i, status.MPI_SOURCE); > } else { > double d; > MPI_Recv(&d, 1, MPI_REAL, 1, 99, MPI_COMM_WORLD, &status); > printf("Received real '%f' from %d.\n", d, status.MPI_SOURCE); > } > } > } > MPI_Finalize(); > return 0; > } > > This example compiles without any warning with the MPICH-3.0.4 library. > However, running: > > $ mpiexec -n 3 ./example > > leads to the following output: > > Probe matched 4 bytes from source 1. > Received real '0.000000' from 1. > Probe matched 4 bytes from source 0. > Received integer '1' from 0. > > Could someone please let me know what is the problem with my program? I > failed to see a problem there. Thank you. > > Best, > > --Jiri Simsa > > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss > -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

Problem reproducing an example from the MPI standard
by Jiri Simsa 22 Jun '13

22 Jun '13

Hi, I tried implementing the Example 3.17 from the MPI 3.0 specification document using as follows: #include <assert.h> #include <mpi.h> #include <stdio.h> int main(int argc, char *argv[]) { int myrank, size; MPI_Status status; MPI_Init(&argc, &argv ); MPI_Comm_rank(MPI_COMM_WORLD, &myrank); MPI_Comm_size(MPI_COMM_WORLD, &size); assert(size == 3); if (myrank == 0) { /* code for process zero */ int i = 1; MPI_Send(&i, 1, MPI_INTEGER, 2, 99, MPI_COMM_WORLD); } if (myrank == 1) { /* code for process one */ double d = 3.14; MPI_Send(&d, 1, MPI_REAL, 2, 99, MPI_COMM_WORLD); } if (myrank == 2) { /* code for process two */ for (int i = 0; i < 2; i++) { MPI_Probe(MPI_ANY_SOURCE, MPI_ANY_TAG, MPI_COMM_WORLD, &status); printf("Probe matched %lld bytes from source %d.\n", status.count, status.MPI_SOURCE); if (status.MPI_SOURCE == 0) { int i; MPI_Recv(&i, 1, MPI_INTEGER, 0, 99, MPI_COMM_WORLD, &status); printf("Received integer '%d' from %d.\n", i, status.MPI_SOURCE); } else { double d; MPI_Recv(&d, 1, MPI_REAL, 1, 99, MPI_COMM_WORLD, &status); printf("Received real '%f' from %d.\n", d, status.MPI_SOURCE); } } } MPI_Finalize(); return 0; } This example compiles without any warning with the MPICH-3.0.4 library. However, running: $ mpiexec -n 3 ./example leads to the following output: Probe matched 4 bytes from source 1. Received real '0.000000' from 1. Probe matched 4 bytes from source 0. Received integer '1' from 0. Could someone please let me know what is the problem with my program? I failed to see a problem there. Thank you. Best, --Jiri Simsa

1 0

Re: [mpich-discuss] run hello world on multiple server
by Antonio J. Peña 21 Jun '13

21 Jun '13

Hi Sufeng, Can you ping/ssh exactly this name "iocfccd1.aps.anl.gov" from iocfccd3? [1] Antonio On Friday, June 21, 2013 10:51:50 AM Sufeng Niu wrote: Hi, Sorry to bother you guys on this stupid question. last time I re-install OS for all blades to keep them the same version. after I mount, set keyless ssh, the terimnal gives the error below: proxy:0:[email protected][2]] HYDU_sock_connect (./utils/sock/sock.c:174): unable to connect from "iocfccd3.aps.anl.gov[3]" to "iocfccd1.aps.anl.gov[1]" (No route to host) [proxy:0:[email protected][2]] main (./pm/pmiserv/pmip.c:189): unable to connect to server iocfccd1.aps.anl.gov[1] at port 38242 (check for firewalls!) I can ssh from iocfccd1 to iocfccd3 without password. Should I shut down all firewalls on each server? I cannot find out where is the problem. Thank you -- Best Regards, Sufeng Niu ECASP lab, ECE department, Illinois Institute of Technology Tel: 312-731-7219[4] -------- [1] http://iocfccd1.aps.anl.gov [2] mailto:proxy%3A0%[email protected] [3] http://iocfccd3.aps.anl.gov [4] tel:312-731-7219

1 0

run hello world on multiple server
by Sufeng Niu 21 Jun '13

21 Jun '13

Hi, Sorry to bother you guys on this stupid question. last time I re-install OS for all blades to keep them the same version. after I mount, set keyless ssh, the terimnal gives the error below: [proxy:0:[email protected]] HYDU_sock_connect (./utils/sock/sock.c:174): unable to connect from "iocfccd3.aps.anl.gov" to "iocfccd1.aps.anl.gov" (No route to host) [proxy:0:[email protected]] main (./pm/pmiserv/pmip.c:189): unable to connect to server iocfccd1.aps.anl.gov at port 38242 (check for firewalls!) I can ssh from iocfccd1 to iocfccd3 without password. Should I shut down all firewalls on each server? I cannot find out where is the problem. Thank you -- Best Regards, Sufeng Niu ECASP lab, ECE department, Illinois Institute of Technology Tel: 312-731-7219

1 0

make error: 'MPID_STATE_MPI_T_CATEGORY_CHANGED' undeclared
by Yi Gu 20 Jun '13

20 Jun '13

Hi, I try to install MPICH 3.04 with logging enabled. So I configure with commands: ./configure --enable-timing=log --with-logging=rlog --enable-timer-type=gettimeofday When configuring rlog, it said "Makefile.in" could not be found in the following directory /mpich-3.0.4/src/util/logging/rlog/ So I copied the "Makefile.in" in mpich-3.0.4/doc/logging/ to that directory and it configured successfully. However, when I tried to make it, there is an error CC src/mpi_t/cat_changed.lo src/mpi_t/cat_changed.c: In function 'PMPI_T_category_changed': src/mpi_t/cat_changed.c:104:5: error: 'MPID_STATE_MPI_T_CATEGORY_CHANGED' undeclared (first use in this function) src/mpi_t/cat_changed.c:104:5: note: each undeclared identifier is reported only once for each function it appears in make[2]: *** [src/mpi_t/cat_changed.lo] Error 1 So how could I solve this problem? Thanks Yi

1 0

Re: [mpich-discuss] Starting processes in root on multiple machines
by Gus Correa 19 Jun '13

19 Jun '13

On 06/19/2013 08:14 AM, Pavan Balaji wrote: > > On 06/19/2013 05:13 AM, Lee, Eibhlin wrote: >> Jim and Pavan will be glad to hear I have now switched to Hydra. I am >> able to run cpi as user on multiple machines. I am also able to run cpi >> as root on one machine. However, I am having difficulty setting up ssh >> properly so that I can run cpi as root on multiple machines. > > Excellent. Once you login as root, can you ssh into the other nodes? > You'll need to set that up first. > >> How can I force a process to run as root on another machine? > > Once the above ssh is setup, you don't need to do anything special. Just > login as root and run your application. > > -- Pavan > I would caution that passwordless ssh for root is not a secure thing. It may be alright if you keep your cluster off the Internet. Maybe the cluster is dedicated to run the MPI A/D converter program? Some people/sites disable remote root logins (or any direct root logins). So, Eibhlin may want to check if any of these restrictive mechanisms exist in his cluster's nodes: http://www.centos.org/docs/4/4.5/Security_Guide/s2-wstation-privileges-noro… I hope this helps, Gus Correa

1 0

Re: [mpich-discuss] Starting processes in root on multiple machines
by Pavan Balaji 19 Jun '13

19 Jun '13

On 06/19/2013 05:13 AM, Lee, Eibhlin wrote: > Jim and Pavan will be glad to hear I have now switched to Hydra. I am > able to run cpi as user on multiple machines. I am also able to run cpi > as root on one machine. However, I am having difficulty setting up ssh > properly so that I can run cpi as root on multiple machines. Excellent. Once you login as root, can you ssh into the other nodes? You'll need to set that up first. > How can I force a process to run as root on another machine? Once the above ssh is setup, you don't need to do anything special. Just login as root and run your application. -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

Starting processes in root on multiple machines
by Lee, Eibhlin 19 Jun '13

19 Jun '13

Hello all, Jim and Pavan will be glad to hear I have now switched to Hydra. I am able to run cpi as user on multiple machines. I am also able to run cpi as root on one machine. However, I am having difficulty setting up ssh properly so that I can run cpi as root on multiple machines. How can I force a process to run as root on another machine? Regards, Eibhlin

1 0

Re: [mpich-discuss] difference between ADIOI_xxx_ReadDone and ADIOI_xxx_ReadComplete
by Matthieu Dorier 17 Jun '13

17 Jun '13

Mmm from the paper you send, I seem to be *almost* correct. ReadDone is just to check without blocking. Thanks again, Matthieu ----- Mail original ----- > De: "Matthieu Dorier" <matthieu.dorier(a)irisa.fr> > À: discuss(a)mpich.org > Envoyé: Lundi 17 Juin 2013 16:11:24 > Objet: Re: [mpich-discuss] difference between ADIOI_xxx_ReadDone and ADIOI_xxx_ReadComplete > > Thanks Rajeev! > > From reading further the code, I understand that the ReadDone > function is called automatically after a read is finished (a sort of > callback, somehow?) while the Complete function would be called when > the user waits on a request corresponding to an asynchronous I/O > operation. Am I correct? > > Matthieu > > ----- Mail original ----- > > De: "Rajeev Thakur" <thakur(a)mcs.anl.gov> > > À: discuss(a)mpich.org > > Envoyé: Lundi 17 Juin 2013 15:22:18 > > Objet: Re: [mpich-discuss] difference between ADIOI_xxx_ReadDone > > and ADIOI_xxx_ReadComplete > > > > If you are looking for 17 year old documentation, see below. But it > > has the answer to your question :-). > > > > Rajeev Thakur, William Gropp, and Ewing Lusk, "An Abstract-Device > > Interface for Implementing Portable Parallel-I/O Interfaces," in > > Proc. of the 6th Symposium on the Frontiers of Massively Parallel > > Computation, October 1996, pp. 180-187. > > http://www.mcs.anl.gov/~thakur/papers/adio.pdf > > > > > > On Jun 17, 2013, at 2:54 PM, Matthieu Dorier wrote: > > > > > Hi, > > > > > > I'm trying to implement an ADIO backend and I'd like to know what > > > is the difference between ADIOI_xxx_ReadDone and > > > ADIOI_xxx_ReadComplete (and respectively for Write)? When are > > > these functions called? (I'm working with mpich 3.0.4) > > > More generally is there somewhere a documentation explaining when > > > and how each of the ADIOI_ functions are called? > > > Thank you, > > > > > > Matthieu Dorier > > > PhD student at ENS Cachan Brittany and IRISA > > > http://people.irisa.fr/Matthieu.Dorier > > > _______________________________________________ > > > 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] difference between ADIOI_xxx_ReadDone and ADIOI_xxx_ReadComplete
by Matthieu Dorier 17 Jun '13

17 Jun '13

Thanks Rajeev! >From reading further the code, I understand that the ReadDone function is called automatically after a read is finished (a sort of callback, somehow?) while the Complete function would be called when the user waits on a request corresponding to an asynchronous I/O operation. Am I correct? Matthieu ----- Mail original ----- > De: "Rajeev Thakur" <thakur(a)mcs.anl.gov> > À: discuss(a)mpich.org > Envoyé: Lundi 17 Juin 2013 15:22:18 > Objet: Re: [mpich-discuss] difference between ADIOI_xxx_ReadDone and ADIOI_xxx_ReadComplete > > If you are looking for 17 year old documentation, see below. But it > has the answer to your question :-). > > Rajeev Thakur, William Gropp, and Ewing Lusk, "An Abstract-Device > Interface for Implementing Portable Parallel-I/O Interfaces," in > Proc. of the 6th Symposium on the Frontiers of Massively Parallel > Computation, October 1996, pp. 180-187. > http://www.mcs.anl.gov/~thakur/papers/adio.pdf > > > On Jun 17, 2013, at 2:54 PM, Matthieu Dorier wrote: > > > Hi, > > > > I'm trying to implement an ADIO backend and I'd like to know what > > is the difference between ADIOI_xxx_ReadDone and > > ADIOI_xxx_ReadComplete (and respectively for Write)? When are > > these functions called? (I'm working with mpich 3.0.4) > > More generally is there somewhere a documentation explaining when > > and how each of the ADIOI_ functions are called? > > Thank you, > > > > Matthieu Dorier > > PhD student at ENS Cachan Brittany and IRISA > > http://people.irisa.fr/Matthieu.Dorier > > _______________________________________________ > > 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