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

Re: [mpich-discuss] hydra crashes with high number of processes
by Antonio J. Peña 24 Jul '13

24 Jul '13

Hi Thomas, Thanks for reporting. Could you please give a try with the current development version and let us know if you still experience the segfault? git clone git://git.mpich.org/mpich.git[1] OR alternatively http://git.mpich.org/mpich.git[2] Thanks, Antonio On Wednesday, July 24, 2013 03:01:30 PM Thomas Ropars wrote: > Hi, > > I'm working with mpich 3.0.4 and I get a segfault in Hydra when I try to > run an application on a large number of processes (8192). > > I simply run the following command: > mpirun -f ~/machine_list -n 8192 my_exec_file > > I tried to run mpirun in valgrind to identify the problem and here is > the output: > Invalid read of size 1 > ==44266== at 0x4A077F2: __GI_strlen (mc_replace_strmem.c:284) > ==44266== by 0x3D774802B5: strdup (in /lib64/libc-2.12.so) > ==44266== by 0x40EBA5: HYD_pmcd_pmi_fill_in_exec_launch_info > (pmiserv_utils.\ > c:375) > ==44266== by 0x40A5C2: HYD_pmci_launch_procs (pmiserv_pmci.c:121) > ==44266== by 0x403A1E: main (mpiexec.c:326) > ==44266== Address 0x0 is not stack'd, malloc'd or (recently) free'd > > If I try to run on a smaller number of processes (eg 512), everything > works fine. > > Any suggestion to solve the problem? > > Thomas > > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss -- Antonio J. Peña Postdoctoral Appointee Mathematics and Computer Science Division Argonne National Laboratory 9700 South Cass Avenue, Bldg. 240, Of. 3148 Argonne, IL 60439-4847 (+1) 630-252-7928 apenya(a)mcs.anl.gov www.mcs.anl.gov/~apenya -------- [1] git://git.mpich.org/mpich.git [2] http://git.mpich.org/mpich.git

1 0

Re: [mpich-discuss] availability of an implementation of MPI 3 on production platforms
by Matthieu Dorier 24 Jul '13

24 Jul '13

Thanks for your reply. I am, indeed, specifically interested in the one-sided part :-) I'll send a mail to the MPI forum. Best, Matthieu ----- Mail original ----- > De: "Jeff Hammond" <jeff.science(a)gmail.com> > À: discuss(a)mpich.org > Envoyé: Mercredi 24 Juillet 2013 09:16:56 > Objet: Re: [mpich-discuss] availability of an implementation of MPI 3 on production platforms > > If you post this to the MPI Forum list, Rich Graham should be able to > share his slide deck of the ETA for MPI-3 for essentially all > implementations. > > MPICH-3 supports all of MPI-3 as far as I know. I think MVAPICH2-1.9 > is derived from it but I have not verified that everything in MPI-3 > is > working (I have no reason to doubt this). > > Word on the street is that Cray deployed their latest MPI > implementation based upon MPICH-3, hence it should support MPI-3 as > well. > > The MPICH test suite exercises many aspects of MPI-3 and can be used > to verify other implementations. > > In the event you're interested in RMA, I have been using most of > MPI-3 > RMA from MPICH-3 regularly. > > Best, > > Jeff > > On Wed, Jul 24, 2013 at 9:59 AM, Matthieu Dorier > <matthieu.dorier(a)irisa.fr> wrote: > > Hi, > > > > This is a non-technical question: > > > > The MPI 3 standard includes some very nice features, which I'm > > impatient to > > leverage in my projects, but being still recent, most platforms > > today (at > > least all production platforms I have access to) do not provide an > > implementation of it. I'm curious when do you think we will be able > > to say > > that a majority of platforms provide MPI 3, and that a code that > > using MPI 3 > > can be seen as portable? (how long did it take between the release > > of the > > MPI 2 standard and its availability on most platforms?) Are you > > aware of > > platforms that already provide it? And what is the status of the > > implementation in Mpich? (are there functions still to be > > implemented?) > > > > Thanks > > > > 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 > > > > -- > Jeff Hammond > jeff.science(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] availability of an implementation of MPI 3 on production platforms
by Jeff Hammond 24 Jul '13

24 Jul '13

If you post this to the MPI Forum list, Rich Graham should be able to share his slide deck of the ETA for MPI-3 for essentially all implementations. MPICH-3 supports all of MPI-3 as far as I know. I think MVAPICH2-1.9 is derived from it but I have not verified that everything in MPI-3 is working (I have no reason to doubt this). Word on the street is that Cray deployed their latest MPI implementation based upon MPICH-3, hence it should support MPI-3 as well. The MPICH test suite exercises many aspects of MPI-3 and can be used to verify other implementations. In the event you're interested in RMA, I have been using most of MPI-3 RMA from MPICH-3 regularly. Best, Jeff On Wed, Jul 24, 2013 at 9:59 AM, Matthieu Dorier <matthieu.dorier(a)irisa.fr> wrote: > Hi, > > This is a non-technical question: > > The MPI 3 standard includes some very nice features, which I'm impatient to > leverage in my projects, but being still recent, most platforms today (at > least all production platforms I have access to) do not provide an > implementation of it. I'm curious when do you think we will be able to say > that a majority of platforms provide MPI 3, and that a code that using MPI 3 > can be seen as portable? (how long did it take between the release of the > MPI 2 standard and its availability on most platforms?) Are you aware of > platforms that already provide it? And what is the status of the > implementation in Mpich? (are there functions still to be implemented?) > > Thanks > > 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 -- Jeff Hammond jeff.science(a)gmail.com

1 0

availability of an implementation of MPI 3 on production platforms
by Matthieu Dorier 24 Jul '13

24 Jul '13

Hi, This is a non-technical question: The MPI 3 standard includes some very nice features, which I'm impatient to leverage in my projects, but being still recent, most platforms today (at least all production platforms I have access to) do not provide an implementation of it. I'm curious when do you think we will be able to say that a majority of platforms provide MPI 3, and that a code that using MPI 3 can be seen as portable? (how long did it take between the release of the MPI 2 standard and its availability on most platforms?) Are you aware of platforms that already provide it? And what is the status of the implementation in Mpich? (are there functions still to be implemented?) Thanks Matthieu Dorier PhD student at ENS Cachan Brittany and IRISA http://people.irisa.fr/Matthieu.Dorier

1 0

hydra crashes with high number of processes
by Thomas Ropars 24 Jul '13

24 Jul '13

Hi, I'm working with mpich 3.0.4 and I get a segfault in Hydra when I try to run an application on a large number of processes (8192). I simply run the following command: mpirun -f ~/machine_list -n 8192 my_exec_file I tried to run mpirun in valgrind to identify the problem and here is the output: Invalid read of size 1 ==44266== at 0x4A077F2: __GI_strlen (mc_replace_strmem.c:284) ==44266== by 0x3D774802B5: strdup (in /lib64/libc-2.12.so) ==44266== by 0x40EBA5: HYD_pmcd_pmi_fill_in_exec_launch_info (pmiserv_utils.\ c:375) ==44266== by 0x40A5C2: HYD_pmci_launch_procs (pmiserv_pmci.c:121) ==44266== by 0x403A1E: main (mpiexec.c:326) ==44266== Address 0x0 is not stack'd, malloc'd or (recently) free'd If I try to run on a smaller number of processes (eg 512), everything works fine. Any suggestion to solve the problem? Thomas

1 0

Re: [mpich-discuss] MPI_Win_free failed
by Pavan Balaji 23 Jul '13

23 Jul '13

Please send a *simple* example which demonstrates the problem. I'm almost certain it's a problem with your application and not with MPI. Hence you going through the exercise of creating a simple example will allow you to figure out what's wrong. -- Pavan On 07/23/2013 06:10 PM, Sufeng Niu wrote: > sure, no problem. but I guess svn maybe easier. since there is some > images file to run it. Can you use svn to clone it? > > it is: https://subversion.xray.aps.anl.gov/detpool/SUFENGNIU/MPI_Compression > > just make, you need two terminals to run it. one terminal type: > ./run.perl main > the second terminal type: ./udp_client 55Fe_run5.tif > > which is that the first terminal send the data, the second terminal > receive it and process it. > > Thank you so much. Due to my limited knowledge I didn't find out where > is the problem. > > once again, thanks a lot! > > Sufeng > > > On Tue, Jul 23, 2013 at 6:00 PM, Pavan Balaji <balaji(a)mcs.anl.gov > <mailto:[email protected]>> wrote: > > > Can you send us a simple program that gives this error? > > -- Pavan > > > On 07/23/2013 05:55 PM, Sufeng Niu wrote: > > Hello, > > I used MPI_Win_free, and it returns error code: 45, Invalid > MPI_Win. I > checked all my code, but I still cannot find out where is the > problem. I > was wondering any one can give a suggestion? > > I defined win as > MPI_Win win; > > in RMA, I create a sub communicator: MPI_FOR_WORLD > MPI_Win_create(image_buff, > 2*image_info->image_size*__sizeof(uint16), > sizeof(uint16), MPI_INFO_NULL, MPI_FOR_WORLD, win); > > In other MPI_FOR_WORLD process: > MPI_Win_create(MPI_BOTTOM, 0, sizeof(uint16), MPI_INFO_NULL, > MPI_FOR_WORLD, win); > > I use MPI_Get, just as normal thing. no mem violation. > > free it as > MPI_Win_free(&win); > > I still cannot figure out why it keeps saying MPI_Win_free > failed. Thank > you very much for any suggestions. > > -- > Best Regards, > Sufeng Niu > ECASP lab, ECE department, Illinois Institute of Technology > Tel: 312-731-7219 <tel:312-731-7219> <tel:312-731-7219 > <tel:312-731-7219>> > > > _________________________________________________ > discuss mailing list discuss(a)mpich.org <mailto:[email protected]> > To manage subscription options or unsubscribe: > https://lists.mpich.org/__mailman/listinfo/discuss > <https://lists.mpich.org/mailman/listinfo/discuss> > > > -- > Pavan Balaji > http://www.mcs.anl.gov/~balaji > > > > > -- > Best Regards, > Sufeng Niu > ECASP lab, ECE department, Illinois Institute of Technology > Tel: 312-731-7219 -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

Re: [mpich-discuss] MPI_Win_free failed
by Sufeng Niu 23 Jul '13

23 Jul '13

sure, no problem. but I guess svn maybe easier. since there is some images file to run it. Can you use svn to clone it? it is: https://subversion.xray.aps.anl.gov/detpool/SUFENGNIU/MPI_Compression just make, you need two terminals to run it. one terminal type: ./run.perl main the second terminal type: ./udp_client 55Fe_run5.tif which is that the first terminal send the data, the second terminal receive it and process it. Thank you so much. Due to my limited knowledge I didn't find out where is the problem. once again, thanks a lot! Sufeng On Tue, Jul 23, 2013 at 6:00 PM, Pavan Balaji <balaji(a)mcs.anl.gov> wrote: > > Can you send us a simple program that gives this error? > > -- Pavan > > > On 07/23/2013 05:55 PM, Sufeng Niu wrote: > >> Hello, >> >> I used MPI_Win_free, and it returns error code: 45, Invalid MPI_Win. I >> checked all my code, but I still cannot find out where is the problem. I >> was wondering any one can give a suggestion? >> >> I defined win as >> MPI_Win win; >> >> in RMA, I create a sub communicator: MPI_FOR_WORLD >> MPI_Win_create(image_buff, 2*image_info->image_size***sizeof(uint16), >> sizeof(uint16), MPI_INFO_NULL, MPI_FOR_WORLD, win); >> >> In other MPI_FOR_WORLD process: >> MPI_Win_create(MPI_BOTTOM, 0, sizeof(uint16), MPI_INFO_NULL, >> MPI_FOR_WORLD, win); >> >> I use MPI_Get, just as normal thing. no mem violation. >> >> free it as >> MPI_Win_free(&win); >> >> I still cannot figure out why it keeps saying MPI_Win_free failed. Thank >> you very much for any suggestions. >> >> -- >> Best Regards, >> Sufeng Niu >> ECASP lab, ECE department, Illinois Institute of Technology >> Tel: 312-731-7219 <tel:312-731-7219> >> >> >> ______________________________**_________________ >> discuss mailing list discuss(a)mpich.org >> To manage subscription options or unsubscribe: >> https://lists.mpich.org/**mailman/listinfo/discuss<https://lists.mpich.org/mailman/listinfo/discuss> >> >> > -- > Pavan Balaji > http://www.mcs.anl.gov/~balaji > -- Best Regards, Sufeng Niu ECASP lab, ECE department, Illinois Institute of Technology Tel: 312-731-7219

1 0

Re: [mpich-discuss] MPI_Win_free failed
by Pavan Balaji 23 Jul '13

23 Jul '13

Can you send us a simple program that gives this error? -- Pavan On 07/23/2013 05:55 PM, Sufeng Niu wrote: > Hello, > > I used MPI_Win_free, and it returns error code: 45, Invalid MPI_Win. I > checked all my code, but I still cannot find out where is the problem. I > was wondering any one can give a suggestion? > > I defined win as > MPI_Win win; > > in RMA, I create a sub communicator: MPI_FOR_WORLD > MPI_Win_create(image_buff, 2*image_info->image_size*sizeof(uint16), > sizeof(uint16), MPI_INFO_NULL, MPI_FOR_WORLD, win); > > In other MPI_FOR_WORLD process: > MPI_Win_create(MPI_BOTTOM, 0, sizeof(uint16), MPI_INFO_NULL, > MPI_FOR_WORLD, win); > > I use MPI_Get, just as normal thing. no mem violation. > > free it as > MPI_Win_free(&win); > > I still cannot figure out why it keeps saying MPI_Win_free failed. Thank > you very much for any suggestions. > > -- > Best Regards, > Sufeng Niu > ECASP lab, ECE department, Illinois Institute of Technology > Tel: 312-731-7219 <tel:312-731-7219> > > > _______________________________________________ > 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

MPI_Win_free failed
by Sufeng Niu 23 Jul '13

23 Jul '13

Hello, I used MPI_Win_free, and it returns error code: 45, Invalid MPI_Win. I checked all my code, but I still cannot find out where is the problem. I was wondering any one can give a suggestion? I defined win as MPI_Win win; in RMA, I create a sub communicator: MPI_FOR_WORLD MPI_Win_create(image_buff, 2*image_info->image_size*sizeof(uint16), sizeof(uint16), MPI_INFO_NULL, MPI_FOR_WORLD, win); In other MPI_FOR_WORLD process: MPI_Win_create(MPI_BOTTOM, 0, sizeof(uint16), MPI_INFO_NULL, MPI_FOR_WORLD, win); I use MPI_Get, just as normal thing. no mem violation. free it as MPI_Win_free(&win); I still cannot figure out why it keeps saying MPI_Win_free failed. Thank you very much for any suggestions. -- Best Regards, Sufeng Niu ECASP lab, ECE department, Illinois Institute of Technology Tel: 312-731-7219

1 0

Re: [mpich-discuss] Question on MPICH rpm for redat 5.6
by Reuti 22 Jul '13

22 Jul '13

Hi, Am 22.07.2013 um 11:50 schrieb McLeish, Allan: > I’m currently having some issues installing MPICH on to our RHEL Linux system due to missing dependencies. > > I want to note that we use Redhat 5.6 and we don’t use YUM so our install is through the use of rpm’s. > > I initially tried installing the latest version of MPICH “mpich2-1.5-3.fc20.x86_64.rpm” but multiple dependencies where required so I tried going to an earliest version of MPICH available in the downloads “mpich2-1.0.8-3.fc11.x86_64.rpm” but this also was missing multiple dependencies. > > MPICH was a 3rd party app required for another program that we were requiring so I contacted there support and they suggested we used MPICH 1.4.1 as it worked the best with there program. Instead of running into these dependencies it might be faster to compile MPICH on your own in the required version. Older version's source you can still get from: http://www.mpich.org/static/downloads/ > When I try to run the RPM for 1.4.1 is asks for multiple dependencies and each of those dependencies have dependencies, is there a tar with all the required dependancies for redhat 5.6 or are there any recommendations on how to go about this? > > > rpm -ivh mpich2-1.4.1p1-9.fc18.x86_64.rpm > error: Failed dependencies: > /usr/bin/bash is needed by mpich2-1.4.1p1-9.fc18.x86_64 > environment-modules is needed by mpich2-1.4.1p1-9.fc18.x86_64 > libc.so.6(GLIBC_2.14)(64bit) is needed by mpich2-1.4.1p1-9.fc18.x86_64 > libc.so.6(GLIBC_2.15)(64bit) is needed by mpich2-1.4.1p1-9.fc18.x86_64 These depends on the version of GLIBC the RPM was created on. If it's compiled directly on the target system it might be better suited. -- Reuti > libc.so.6(GLIBC_2.7)(64bit) is needed by mpich2-1.4.1p1-9.fc18.x86_64 > libhwloc.so.5()(64bit) is needed by mpich2-1.4.1p1-9.fc18.x86_64 > rpmlib(FileDigests) <= 4.6.0-1 is needed by mpich2-1.4.1p1-9.fc18.x86_64 > rpmlib(PayloadIsXz) <= 5.2-1 is needed by mpich2-1.4.1p1-9.fc18.x86_64 > > > Many Thanks > > Allan > > Maersk Oil North Sea UK Limited, registered in England and Wales No. 03682299. Registered office Maersk House, Braham Street, London E1 8EP. > This e-mail and any files transmitted with it are confidential and intended solely for the use of the individual or entity to which they are addressed. If you have received this e-mail in error please notify the system manager at hotline(a)maerskoil.com. > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss

1 0