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

Re: [mpich-discuss] MPI_Comm_Spawn causing zombies of hydra_pmi_proxy
by Silvan Brändli 21 Mar '13

21 Mar '13

PS: Just some additional information on compilers: so far I used icpc, when changing it to gcc-4.7.2 I still get the zombies. The only change in the linked libraries is an additional libdl.so.2 => /lib64/libdl.so.2 Any hint / idea? Best regards Silvan On 19.03.2013 10:47, Silvan Brï¿½ndli wrote: > Dear Pavan, > > in the attached example I use mpich3, but still I get the zombies. Is > there something wrong with > - my MPI function calls? (disconnect, finalize) > - my linked libraries? (see below) > > Thanks in advance! > Best regards > Silvan > > ldd_hi > linux-vdso.so.1 (0x00007fffa01ff000) > libmpichcxx.so.10 => /opt/mpich3/lib/libmpichcxx.so.10 > (0x00007fdfec0e4000) > libmpich.so.10 => /opt/mpich3/lib/libmpich.so.10 (0x00007fdfebc6f000) > libopa.so.1 => /opt/mpich3/lib/libopa.so.1 (0x00007fdfeba6d000) > libmpl.so.1 => /opt/mpich3/lib/libmpl.so.1 (0x00007fdfeb868000) > libaio.so.1 => /lib64/libaio.so.1 (0x00007fdfeb666000) > libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fdfeb449000) > libm.so.6 => /lib64/libm.so.6 (0x00007fdfeb14e000) > libstdc++.so.6 => > /usr/lib/gcc/x86_64-pc-linux-gnu/4.7.2/libstdc++.so.6 (0x00007fdfeae47000) > libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x00007fdfeac30000) > libc.so.6 => /lib64/libc.so.6 (0x00007fdfea882000) > libdl.so.2 => /lib64/libdl.so.2 (0x00007fdfea67e000) > libgfortran.so.3 => > /usr/lib/gcc/x86_64-pc-linux-gnu/4.7.2/libgfortran.so.3 > (0x00007fdfea361000) > libquadmath.so.0 => > /usr/lib/gcc/x86_64-pc-linux-gnu/4.7.2/libquadmath.so.0 > (0x00007fdfea12b000) > /lib64/ld-linux-x86-64.so.2 (0x00007fdfec308000) > librt.so.1 => /lib64/librt.so.1 (0x00007fdfe9f23000) > ldd_main > linux-vdso.so.1 (0x00007fffeb1ff000) > libmpichcxx.so.10 => /opt/mpich3/lib/libmpichcxx.so.10 > (0x00007f6fc797e000) > libmpich.so.10 => /opt/mpich3/lib/libmpich.so.10 (0x00007f6fc7509000) > libopa.so.1 => /opt/mpich3/lib/libopa.so.1 (0x00007f6fc7307000) > libmpl.so.1 => /opt/mpich3/lib/libmpl.so.1 (0x00007f6fc7102000) > libaio.so.1 => /lib64/libaio.so.1 (0x00007f6fc6f00000) > libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f6fc6ce3000) > libm.so.6 => /lib64/libm.so.6 (0x00007f6fc69e8000) > libstdc++.so.6 => > /usr/lib/gcc/x86_64-pc-linux-gnu/4.7.2/libstdc++.so.6 (0x00007f6fc66e1000) > libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x00007f6fc64ca000) > libc.so.6 => /lib64/libc.so.6 (0x00007f6fc611c000) > libdl.so.2 => /lib64/libdl.so.2 (0x00007f6fc5f18000) > libgfortran.so.3 => > /usr/lib/gcc/x86_64-pc-linux-gnu/4.7.2/libgfortran.so.3 > (0x00007f6fc5bfb000) > libquadmath.so.0 => > /usr/lib/gcc/x86_64-pc-linux-gnu/4.7.2/libquadmath.so.0 > (0x00007f6fc59c5000) > /lib64/ld-linux-x86-64.so.2 (0x00007f6fc7ba2000) > librt.so.1 => /lib64/librt.so.1 (0x00007f6fc57bd000) > > On 07.03.2013 17:23, Pavan Balaji wrote: >> >> I just tried this with the mpich master and it seems to work correctly, >> and there are no zombie processes (though I reduced the number of >> iterations to 10000, instead of 200000). This was a problem in mpich >> once upon a time, but that was a few years ago. Are you using the >> latest version of mpich (3.0.2)? >> >> -- Pavan >> >> On 03/07/2013 09:44 AM US Central Time, Silvan Brï¿½ndli wrote: >>> PS: The attached programs are a simplification of my code. They >>> reproduce the zombie problem. Waiting for the 32k zombies takes a >>> while... but I expect the same behaviour as with my original code. >>> >>> Am I missing something when finishing the called program? I just use >>> MPI_Comm_disconnect and MPI_Finalize. >>> >>> Best regards >>> Silvan >>> >>> main.cpp >>> >>> #include <mpi.h>; >>> >>> int main(int argc, char *argv[]) >>> { >>> int myrank; >>> int spawnerror; >>> int value = 123; >>> void *buf = &value; >>> MPI_Comm child_comm; >>> >>> if (MPI_Init(&argc,&argv)!=MPI_SUCCESS) >>> { >>> printf("MPI_Init failed"); >>> } >>> >>> MPI_Comm_rank(MPI_COMM_WORLD, &myrank); >>> >>> char* hiargv[] = {"23",NULL}; >>> for(int i = 1; i <= 200000; i++) >>> { >>> value = i; >>> printf("Main before spawn %d\n",i); >>> MPI_Comm_spawn("./hi",hiargv, 1, MPI_INFO_NULL, myrank, >>> MPI_COMM_SELF, &child_comm, &spawnerror); >>> MPI_Send(buf, 1, MPI_INTEGER, 0, 1, child_comm); >>> MPI_Comm_disconnect(&child_comm); >>> } >>> >>> MPI_Finalize(); >>> return 0; >>> } >>> >>> hi.cpp: >>> >>> #include <mpi.h>; >>> >>> int main(int argc, char** argv) { >>> MPI_Comm parent; >>> MPI_Status status; >>> int err; >>> int value = -1; >>> void* buf= &value; >>> >>> if (MPI_Init(&argc,&argv)!=MPI_SUCCESS) >>> { >>> printf("MPI_Init failed"); >>> } >>> MPI_Comm_get_parent(&parent); >>> if (parent == MPI_COMM_NULL) printf("No parent!"); >>> >>> MPI_Recv(buf, 1, MPI_INTEGER, 0, MPI_ANY_TAG, parent, &status); >>> MPI_Comm_disconnect(&parent); >>> err = MPI_Finalize(); >>> printf("hi finalized %d %d \n",err, value); >>> return 0; >>> } >>> >>> >>> >>> On 07.03.2013 12:38, Silvan Brï¿½ndli wrote: >>>> Dear all, >>>> >>>> again I have a question related to spawning processes. I understand the >>>> situation as follows: >>>> >>>> My program A spawns program B. Program B spawns program C1, C2 ... >>>> C10000 ... >>>> Program Cx terminates correctly before Cx+1 is called, however >>>> returning >>>> 1 to mpiexec. To handle this I use the workaround as described in >>>> http://lists.mpich.org/pipermail/discuss/2013-February/000429.html >>>> >>>> Now it looks like with every Spawn a "hydra_pmi_proxy" is started, the >>>> calling program is mpiexec. When the program Cx is finished this >>>> "hydra_pmi_proxy" remains as a zombie until the programs A, B and >>>> mpiexec are finished. When approx. 32k of those "hydra_pmi_proxy" exist >>>> I get some problems (too many processes or something similar). >>>> >>>> What can I do to finish "hydra_pmi_proxy" while my programs A, B and >>>> mpiexec are still running? >>>> >>>> I'm glad about every hint. >>>> >>>> Best regards >>>> Silvan >>>> >>> >>> >>> >>> >>> _______________________________________________ >>> 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 > -- Dipl.-Ing. Silvan Brï¿½ndli Numerische Strukturanalyse mit Anwendungen in der Schiffstechnik (M-10) Technische Universitï¿½t Hamburg-Harburg Schwarzenbergstraï¿½e 95c 21073 Hamburg Tel. : +49 (0)40 42878 - 6187 Fax. : +49 (0)40 42878 - 6090 e-mail: silvan.braendli(a)tuhh.de www : http://www.tuhh.de/skf 5th GACM Colloquium on Computational Mechanics http://www.tu-harburg.de/gacm2013

1 0

Re: [mpich-discuss] Query regarding structure of dynamic array derived datatype
by Mahesh Doijade 21 Mar '13

21 Mar '13

Hi, Thanks for your reply, after looking at various MPI derived datatype creation functions other than MPI_Pack according to me option can probably be of using MPI_Type_create_struct but then for each structure element I have to create a separate derived datatype as the block lengths will be different due to dynamic array present in them, so again the extra memory will be required now for type-maps of these datatypes. Is the approach of MPI_Type_create_struct mentioned above will be better than MPI_Pack in this case? also is there any other optimal way to do this? -- Regards, -- Mahesh Doijade On Thu, Mar 21, 2013 at 2:36 AM, Pavan Balaji <balaji(a)mcs.anl.gov> wrote: > > Please look through the derived datatype constructs in MPI. Google can > help you find some tutorials. > > -- Pavan > > On 03/19/2013 06:12 AM US Central Time, Mahesh Doijade wrote: > > Hi, > > I have a struct as - > > struct A > > { > > int size; > > int *lptr; > > }; > > And then I have array of such structure, I want to only send > > dynamic array present in each of the structs to other process. > > One of the options is to do it with MPI_Pack but then I have read it is > > not so good from performance point of view. > > What can be the other opt > > > > -- > > > > Regards, > > -- Mahesh Doijade > > > > > > > > > > > > _______________________________________________ > > 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

Re: [mpich-discuss] Re : MPI I/O vs Sequential I/O
by Rajeev Thakur 20 Mar '13

20 Mar '13

megabytes, gigabytes. On Mar 20, 2013, at 6:13 PM, Muhammad Zulfikar Handana wrote: > > exactly how big the files are great for use on a comparison of MPI I / O? > > thank a lot for your attention > > handana > > fwrite() does buffered I/O, which will perform better for small writes. A better comparison would be to use write(), which is what MPI-IO uses internally. > > Rajeev > > On Mar 18, 2013, at 3:49 PM, Muhammad Zulfikar Handana wrote: > > > hello, i have some questions about MPI-I/O. > > > > 1) when i trying execute my program using MPI-I/O. why when iam execute > > with 1 proses, execute time more good than when iam execute with 2 proses. > > am i wrong? this is my source code. > > MPI_File_open (MPI_COMM_WORLD, file_akhir, MPI_MODE_CREATE | > > MPI_MODE_WRONLY, MPI_INFO_NULL, &hasil); > > int jumlah1 = (((block-1)*16)+pad); > > int jumlah = jumlah1 / size; > > > > > > MPI_File_set_view(hasil, rank * jumlah * > > sizeof(char),MPI_CHAR, MPI_CHAR, "native", MPI_INFO_NULL); > > MPI_File_write (hasil, output, jumlah1, MPI_CHAR, > > MPI_STATUS_IGNORE); > > > > MPI_File_close(&hasil); > > > > 2) and after that i have compared that source using MPI - I/O and with > > sequential I/O, apparently, sequential i / o is better than the parallel i > > / o. how it could be happen? this is my sequential source. > > if (rank==0) { > > > > hasil = fopen (file_akhir, "wb"); > > int test = (((block-1)*16)+pad); > > fwrite (output, sizeof(char), test, hasil ); > > fclose (hasil); > > } > > > > thanks a lot for your attention > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re : MPI I/O vs Sequential I/O
by Muhammad Zulfikar Handana 20 Mar '13

20 Mar '13

exactly how big the files are great for use on a comparison of MPI I / O? thank a lot for your attention handana fwrite() does buffered I/O, which will perform better for small writes. A better comparison would be to use write(), which is what MPI-IO uses internally. Rajeev On Mar 18, 2013, at 3:49 PM, Muhammad Zulfikar Handana wrote: > hello, i have some questions about MPI-I/O. > > 1) when i trying execute my program using MPI-I/O. why when iam execute > with 1 proses, execute time more good than when iam execute with 2 proses. > am i wrong? this is my source code. > MPI_File_open (MPI_COMM_WORLD, file_akhir, MPI_MODE_CREATE | > MPI_MODE_WRONLY, MPI_INFO_NULL, &hasil); > int jumlah1 = (((block-1)*16)+pad); > int jumlah = jumlah1 / size; > > > MPI_File_set_view(hasil, rank * jumlah * > sizeof(char),MPI_CHAR, MPI_CHAR, "native", MPI_INFO_NULL); > MPI_File_write (hasil, output, jumlah1, MPI_CHAR, > MPI_STATUS_IGNORE); > > MPI_File_close(&hasil); > > 2) and after that i have compared that source using MPI - I/O and with > sequential I/O, apparently, sequential i / o is better than the parallel i > / o. how it could be happen? this is my sequential source. > if (rank==0) { > > hasil = fopen (file_akhir, "wb"); > int test = (((block-1)*16)+pad); > fwrite (output, sizeof(char), test, hasil ); > fclose (hasil); > } > > thanks a lot for your attention

1 0

Re: [mpich-discuss] Query regarding structure of dynamic array derived datatype
by Pavan Balaji 20 Mar '13

20 Mar '13

Please look through the derived datatype constructs in MPI. Google can help you find some tutorials. -- Pavan On 03/19/2013 06:12 AM US Central Time, Mahesh Doijade wrote: > Hi, > I have a struct as - > struct A > { > int size; > int *lptr; > }; > And then I have array of such structure, I want to only send > dynamic array present in each of the structs to other process. > One of the options is to do it with MPI_Pack but then I have read it is > not so good from performance point of view. > What can be the other opt > > -- > > Regards, > -- Mahesh Doijade > MTech - II (CSE) , > Walchand College of Engineering, Sangli > > > > > > _______________________________________________ > 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

Re: [mpich-discuss] MPI on Intel MIC
by Pavan Balaji 20 Mar '13

20 Mar '13

Martin, For the MIC build, you'll just need to pass CFLAGS=-mmic. You didn't give enough information below to say what went wrong. -- Pavan On 03/19/2013 02:26 AM US Central Time, Martin Rose wrote: > Dear all, > > I want to run an MPI program on an Intel Xeon Phi accelerator (MIC). I > want to use an open source MPI library but I have not found any > information about how to compile the MPI library for MIC. > > The link below leads to a paper about MVAPICH2 on MIC but the paper does > not talk about compilation of the library. > http://nowlab.cse.ohio-state.edu/publications/conf-papers/2012/potluri-tihp… > > > I have tried to compile mpich-3.0.2 with icc anf ifort using a -dmmic as > dummy flag. After running config, the dummy flag is replaced by -mmic. > This got me pretty far but I do not have libnuma for MIC so the > compilation could not complete. > > The announcement of MPICH2-1.5 states that the MIC architecture is > supported. > http://www.mpich.org/2012/10/08/mpich2-1-5-released-oct-8-2012/ > > Can anyone describe the build process to run a program compiled with > MPICH on Intel MIC? > > Greetings, > Martin > _______________________________________________ > 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

Re: [mpich-discuss] running out of communicators
by Jim Dinan 20 Mar '13

20 Mar '13

Hi Ryan, If you set a breakpoint at MPIR_Err_return_comm, you should be able to see the rest of the call stack where the error occurred. ~Jim. On 3/20/13 3:08 PM, Ryan Crocker wrote: > So i ran my code through totalview and it didn't pick up on anything, but i still crash with the same error. I'm starting to think it's one of my third party programs. I'm using Hypre as my Poission solver. Does anybody know if that could be causing the error? > > -Ryan > On Mar 19, 2013, at 8:18 PM, Jim Dinan wrote: > >> I don't recall the exact limit on the number of communicators, but is certainly much more than 21. Any chance that you're forgetting to close one of those files somewhere? Can you attach a debugger to confirm that the error is happening in MPI_File_open? >> >> ~Jim. >> >> On 3/19/13 8:50 PM, Ryan Crocker wrote: >>> I'm opening 21 files and i do have the close calls after they are finished being written. >>> >>> On Mar 19, 2013, at 6:29 PM, Jim Dinan wrote: >>> >>>> Hi Ryan, >>>> >>>> MPI_File_open does call MPI_Comm_dup internally. How many files are you opening? And are you closing them when you're finished with them? >>>> >>>> ~Jim. >>>> >>>> On 3/19/13 8:20 PM, Ryan Crocker wrote: >>>>> Hi Jim, >>>>> >>>>> That's the thing, i'm not sure where they are being created. I initialize my MPI environment, then that program never gets called again. The only think i can think of is that it's something to do with my MPI IO, but after every open file, i just double checked, there is a corresponding close file call. I also do not call MPI_Comm_dup, unless that is called by another mpi call that i am not aware of. >>>>> >>>>> -Ryan >>>>> >>>>> On Mar 19, 2013, at 6:13 PM, Jim Dinan wrote: >>>>> >>>>>> Hi Ryan, >>>>>> >>>>>> Every time you call MPI_Comm_dup a new communicator is created. Are you ever freeing these, using MPI_Comm_free? >>>>>> >>>>>> Also, what are you trying to achieve by using multiple communicators, and why does it require so many? >>>>>> >>>>>> Best, >>>>>> ~Jim. >>>>>> >>>>>> On 3/19/13 7:57 PM, Ryan Crocker wrote: >>>>>>> I realized i forgot to attach the error: >>>>>>> >>>>>>> Fatal error in PMPI_Comm_dup: Other MPI error, error stack: >>>>>>> PMPI_Comm_dup(176)............: MPI_Comm_dup(comm=0x84000000, new_comm=0x7fff5fbfe9a4) failed >>>>>>> PMPI_Comm_dup(161)............: >>>>>>> MPIR_Comm_dup_impl(55)........: >>>>>>> MPIR_Comm_copy(967)...........: >>>>>>> MPIR_Get_contextid(521).......: >>>>>>> MPIR_Get_contextid_sparse(752): Too many communicators >>>>>>> >>>>>>> Hi all, >>>>>>> >>>>>>> So i can't seem to find the answer to this question. I keep getting the failure "Too many communicators". Could someone explain what calls "use up" communicators, and could i be calling/using more than the default total?. I'm trying to debug my code so i can free up the communicators and not just repeatedly create them. I'm a bit perplexed with this, probably a lack of in depth knowledge, but i only run any of my MPI initializations one time then all my calls with mpi are either sums, max, min, allreduce, alltoall. >>>>>>> >>>>>>> thanks >>>>>>> >>>>>>> Ryan Crocker >>>>>>> University of Vermont, School of Engineering >>>>>>> Mechanical Engineering Department >>>>>>> rcrocker(a)uvm.edu >>>>>>> 315-212-7331 >>>>>>> >>>>>>> _______________________________________________ >>>>>>> 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 >>>>> >>>>> Ryan Crocker >>>>> University of Vermont, School of Engineering >>>>> Mechanical Engineering Department >>>>> rcrocker(a)uvm.edu >>>>> 315-212-7331 >>>>> >>>>> _______________________________________________ >>>>> 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 >>> >>> Ryan Crocker >>> University of Vermont, School of Engineering >>> Mechanical Engineering Department >>> rcrocker(a)uvm.edu >>> 315-212-7331 >>> >>> _______________________________________________ >>> 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 > > Ryan Crocker > University of Vermont, School of Engineering > Mechanical Engineering Department > rcrocker(a)uvm.edu > 315-212-7331 > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss >

1 0

Hypre solvers and MPICH
by Ryan Crocker 20 Mar '13

20 Mar '13

Hi all, I had posted a thread before about getting a "too many communicators" error after a long run. After going through with totalview i didn't see any problems. I am using Hypre in MPI as a Poission solver. I think this is probably causing the error because i need to create and destroy multiple Hypre solver constructs during my run (at almost every iteration) . Does anyone know enough about Hypre coupled with MPICH to know if Hypre frees all the communicators when it destroys the constructs? I looked over the hypre message board but it's pretty dead, i was wondering if i'll have better luck here. Ryan Crocker University of Vermont, School of Engineering Mechanical Engineering Department

1 0

Re: [mpich-discuss] running out of communicators
by Ryan Crocker 20 Mar '13

20 Mar '13

So i ran my code through totalview and it didn't pick up on anything, but i still crash with the same error. I'm starting to think it's one of my third party programs. I'm using Hypre as my Poission solver. Does anybody know if that could be causing the error? -Ryan On Mar 19, 2013, at 8:18 PM, Jim Dinan wrote: > I don't recall the exact limit on the number of communicators, but is certainly much more than 21. Any chance that you're forgetting to close one of those files somewhere? Can you attach a debugger to confirm that the error is happening in MPI_File_open? > > ~Jim. > > On 3/19/13 8:50 PM, Ryan Crocker wrote: >> I'm opening 21 files and i do have the close calls after they are finished being written. >> >> On Mar 19, 2013, at 6:29 PM, Jim Dinan wrote: >> >>> Hi Ryan, >>> >>> MPI_File_open does call MPI_Comm_dup internally. How many files are you opening? And are you closing them when you're finished with them? >>> >>> ~Jim. >>> >>> On 3/19/13 8:20 PM, Ryan Crocker wrote: >>>> Hi Jim, >>>> >>>> That's the thing, i'm not sure where they are being created. I initialize my MPI environment, then that program never gets called again. The only think i can think of is that it's something to do with my MPI IO, but after every open file, i just double checked, there is a corresponding close file call. I also do not call MPI_Comm_dup, unless that is called by another mpi call that i am not aware of. >>>> >>>> -Ryan >>>> >>>> On Mar 19, 2013, at 6:13 PM, Jim Dinan wrote: >>>> >>>>> Hi Ryan, >>>>> >>>>> Every time you call MPI_Comm_dup a new communicator is created. Are you ever freeing these, using MPI_Comm_free? >>>>> >>>>> Also, what are you trying to achieve by using multiple communicators, and why does it require so many? >>>>> >>>>> Best, >>>>> ~Jim. >>>>> >>>>> On 3/19/13 7:57 PM, Ryan Crocker wrote: >>>>>> I realized i forgot to attach the error: >>>>>> >>>>>> Fatal error in PMPI_Comm_dup: Other MPI error, error stack: >>>>>> PMPI_Comm_dup(176)............: MPI_Comm_dup(comm=0x84000000, new_comm=0x7fff5fbfe9a4) failed >>>>>> PMPI_Comm_dup(161)............: >>>>>> MPIR_Comm_dup_impl(55)........: >>>>>> MPIR_Comm_copy(967)...........: >>>>>> MPIR_Get_contextid(521).......: >>>>>> MPIR_Get_contextid_sparse(752): Too many communicators >>>>>> >>>>>> Hi all, >>>>>> >>>>>> So i can't seem to find the answer to this question. I keep getting the failure "Too many communicators". Could someone explain what calls "use up" communicators, and could i be calling/using more than the default total?. I'm trying to debug my code so i can free up the communicators and not just repeatedly create them. I'm a bit perplexed with this, probably a lack of in depth knowledge, but i only run any of my MPI initializations one time then all my calls with mpi are either sums, max, min, allreduce, alltoall. >>>>>> >>>>>> thanks >>>>>> >>>>>> Ryan Crocker >>>>>> University of Vermont, School of Engineering >>>>>> Mechanical Engineering Department >>>>>> rcrocker(a)uvm.edu >>>>>> 315-212-7331 >>>>>> >>>>>> _______________________________________________ >>>>>> 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 >>>> >>>> Ryan Crocker >>>> University of Vermont, School of Engineering >>>> Mechanical Engineering Department >>>> rcrocker(a)uvm.edu >>>> 315-212-7331 >>>> >>>> _______________________________________________ >>>> 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 >> >> Ryan Crocker >> University of Vermont, School of Engineering >> Mechanical Engineering Department >> rcrocker(a)uvm.edu >> 315-212-7331 >> >> _______________________________________________ >> 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 Ryan Crocker University of Vermont, School of Engineering Mechanical Engineering Department rcrocker(a)uvm.edu 315-212-7331

1 0

Re: [mpich-discuss] MPI I/O vs Sequential I/O
by Muhammad Zulfikar Handana 20 Mar '13

20 Mar '13

thanks for your explain here. iam trying wrote into file which have 4.5 MB and 115 MB. i think that large amount data. iam using file JPG and MP4 extension. actually this program about Advanced Encryption Standard (AES), iam gonna try to using MPI I/O for reducing execute time this program. oh what do you mean about "that only recovers at much higher levels of parallelism"? can you explain the details and with example? thanks a lot for your attention > 1) when i trying execute my program using MPI-I/O. why when iam execute > with 1 proses, execute time more good than when iam execute with 2 proses. > am i wrong? this is my source code. In general, with one process doing i/o the file system can cache the heck out of requests. When a second process comes along, the caching needs to be disabled or invalidated, causing a "performance crash" that only recovers at much higher levels of parallelism. Let's look specifically at your code, though: MPI_File_open (MPI_COMM_WORLD, file_akhir, MPI_MODE_CREATE | MPI_MODE_WRONLY, MPI_INFO_NULL, &hasil); int jumlah1 = (((block-1)*16)+pad); int jumlah = jumlah1 / size; MPI_File_set_view(hasil, rank * jumlah * sizeof(char), MPI_CHAR, MPI_CHAR, "native", MPI_INFO_NULL); MPI_File_write (hasil, output, jumlah1, MPI_CHAR, MPI_STATUS_IGNORE); MPI_File_close(&hasil); - Are you writing a large amount of data or a small amount of data here? - What file system are you using? -- Rob Latham Mathematics and Computer Science Division Argonne National Lab, IL USA ------------------------------ _______________________________________________ discuss mailing list discuss(a)mpich.org https://lists.mpich.org/mailman/listinfo/discuss

1 0