May 2016 - discuss - mailman3.cels.anl.gov

Re: [mpich-discuss] MPI installation
by Oden, Lena 11 May '16

11 May '16

According to the attached file, you installed MPICH at /usr/local/ so you should find the binaries at /usr/local/bin, the libs at /usr/local/lib etc. You maybe build MPICH in /usr/local/mpich_install/mpich-3.2 - but you didn’t install it there. Did you use the —prefix configure option? Lena > On May 11, 2016, at 11:34 AM, KEDAR GOVIND BHIDE <kedarbhide(a)iitb.ac.in> wrote: > > <mi.txt> _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

MPI installation
by KEDAR GOVIND BHIDE 11 May '16

11 May '16

I've followed the 7 instructions upto page 4 in the installation manual. https://www.mpich.org/static/downloads/3.2/mpich-3.2-installguide.pdf I've installed mpich at following place: /usr/local/mpich_install/mpich-3.2 I cannot find 'bin' folder inside mpich-3.2 and thus changing 'bashrc' file won't work. Also attached with email is 'mi.txt' generated after installation. OpenMPI is installed in our cluster right now. Command 'which mpcc' shows OpenMPI files. Please help us to install MPI libraries on our clusters. -- Kedar G. Bhide PhD student Mechanical Engg. Dept Indian Institute of Technology,Bombay India +919920296985 _______________________________________________ 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] MPI_Comm_accept() pins cpu core 100%
by Balaji, Pavan 10 May '16

10 May '16

This configuration does not support full MPI-3 and will be deprecated in the future. You'll also lose some performance compared to the default configuration, but, well, you are oversubscribing cores, so I don't think that's as much of a concern in your case. -- Pavan > On May 9, 2016, at 11:08 AM, Dmitriy Lyubimov <dlieu.7(a)gmail.com> wrote: > > ok, thank you. > > What implications does this have? does this reduce the solution capabilities/performance in any specific way? > > for now, i ended up just bumping process priority to the max priority for the duration of accept call, which is still not great but i don't want to mess with the build capabilities or ask people to install a custom build. > > On Mon, May 9, 2016 at 6:35 AM, Balaji, Pavan <balaji(a)anl.gov> wrote: > > For now, the only workaround for this is to configure with --with-device=ch3:sock > > -- Pavan > > > On May 6, 2016, at 5:49 PM, Dmitriy Lyubimov <dlieu.7(a)gmail.com> wrote: > > > > Hi, > > > > I have a situation when a process my start up and execute MPI_Comm_accept(), while waiting for others to join, which may take some time. > > > > I have observed that all processes that wait in MPI_Comm_accept(), pin 1 cpu core exactly 100%. > > > > Which is unfortunate in my situation since if i launch enough processes, i pin the entire machine node very quickly -- and it does do other things that run into a crawl. > > > > Is this a known problem? is there a workaround for this? > > > > I tried on ubuntu 14-04 LTS which installs mpich 3.0.4 by default but i also tested 3.2 Nov/2015 and they both exhibit this very behavior. > > > > Thank you very much in advance. > > -Dmitriy > > _______________________________________________ > > 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] Enable Debugging MPICH
by lightenlee 10 May '16

10 May '16

Hi Halim, you are right, thank you very much. --Ice At 2016-05-09 22:48:39, "Halim Amer" <aamer(a)anl.gov> wrote: >I don't see why you couldn't step into MPI_Get_processor_name, other >than you are linking to the wrong MPI implementation that doesn't have >debug symbols. > >I tested your example with --enable-g=all when building MPICH and it was >sufficient to step into MPI_Get_processor_name. > >--Halim > >On 5/8/16 6:56 AM, 冰 wrote: >> Hi, >> I want to use |gdb| to step into the code to see what's going on inside >> the code, for example my code as below, I set a break at the line >> MPI_Get_processor_name (processor_name, &namelen), I want to step into >> MPI_Get_processor_name, I configure with |--enable-g=all and >> |--enable-fast=O0, but gdb cann't step into MPI_Get_processor_name, it >> steps into fprintf, what should I do? >> >> #include "mpi.h" >> #include <stdio.h> >> int main (int argc, char **argv) >> { >> int myid, numprocs; >> int namelen; >> char processor_name[MPI_MAX_PROCESSOR_NAME]; >> >> MPI_Init (&argc, &argv); >> MPI_Comm_rank (MPI_COMM_WORLD, &myid); >> MPI_Comm_size (MPI_COMM_WORLD, &numprocs); >> MPI_Get_processor_name (processor_name, &namelen); >> fprintf (stderr, "Hello World! Process %d of %d on %s\n", myid, >> numprocs, processor_name); >> MPI_Finalize (); >> return 0; >> } >> >> >> >> >> >> >> _______________________________________________ >> 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] allgatherv
by Halim Amer 09 May '16

09 May '16

Never mind, Rajeev was right; I missed the medium messages with powers of two case. --Halim On 5/9/16 12:45 PM, Halim Amer wrote: > Looking at the code, for long messages the same ring algorithm is used > for long/medium messages regardless of whether the number of processes > is a power of two or not. The comment is wrong. We will fix it, but in > general one shouldn't rely solely on code comments and should look at > the code itself. > > --Halim > www.mcs.anl.gov/~aamer > > On 5/9/16 12:21 PM, Jen B wrote: >> Hello, >> >> I am looking at allgatherv.c from: >> >> git.mpich.org/mpich.git/tree/HEAD:/src/mpi/coll >> <http://git.mpich.org/mpich.git/tree/HEAD:/src/mpi/coll> >> >> It says the following: >> >> "For short or medium-size messages and power-of-two no. of processes, >> we use the recursive doubling algorithm. >> ... >> For long messages or medium-size messages and non-power-of-two >> no. of processes, we use a ring algorithm. >> ..." >> >> Could someone please tell me what algorithm is used when we have "long >> messages and power-of-two no. of processes"? Thanks a lot! >> >> Best, >> Jen >> >> >> _______________________________________________ >> 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] allgatherv
by Halim Amer 09 May '16

09 May '16

Looking at the code, for long messages the same ring algorithm is used for long/medium messages regardless of whether the number of processes is a power of two or not. The comment is wrong. We will fix it, but in general one shouldn't rely solely on code comments and should look at the code itself. --Halim www.mcs.anl.gov/~aamer On 5/9/16 12:21 PM, Jen B wrote: > Hello, > > I am looking at allgatherv.c from: > > git.mpich.org/mpich.git/tree/HEAD:/src/mpi/coll > <http://git.mpich.org/mpich.git/tree/HEAD:/src/mpi/coll> > > It says the following: > > "For short or medium-size messages and power-of-two no. of processes, > we use the recursive doubling algorithm. > ... > For long messages or medium-size messages and non-power-of-two > no. of processes, we use a ring algorithm. > ..." > > Could someone please tell me what algorithm is used when we have "long > messages and power-of-two no. of processes"? Thanks a lot! > > Best, > Jen > > > _______________________________________________ > 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] allgatherv
by Jen B 09 May '16

09 May '16

Oh, I see. Thank you! On Mon, May 9, 2016 at 12:27 PM, Thakur, Rajeev <thakur(a)mcs.anl.gov> wrote: > Read it this way with the parenthesis: > > For long messages or (medium-size messages and non-power-of-two no. of > processes) > > > > On May 9, 2016, at 12:21 PM, Jen B <jenbna(a)gmail.com> wrote: > > > > Hello, > > > > I am looking at allgatherv.c from: > > > > git.mpich.org/mpich.git/tree/HEAD:/src/mpi/coll > > > > It says the following: > > > > "For short or medium-size messages and power-of-two no. of processes, > > we use the recursive doubling algorithm. > > ... > > For long messages or medium-size messages and non-power-of-two > > no. of processes, we use a ring algorithm. > > ..." > > > > Could someone please tell me what algorithm is used when we have "long > messages and power-of-two no. of processes"? Thanks a lot! > > > > Best, > > Jen > > _______________________________________________ > > 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] allgatherv
by Thakur, Rajeev 09 May '16

09 May '16

Read it this way with the parenthesis: For long messages or (medium-size messages and non-power-of-two no. of processes) > On May 9, 2016, at 12:21 PM, Jen B <jenbna(a)gmail.com> wrote: > > Hello, > > I am looking at allgatherv.c from: > > git.mpich.org/mpich.git/tree/HEAD:/src/mpi/coll > > It says the following: > > "For short or medium-size messages and power-of-two no. of processes, > we use the recursive doubling algorithm. > ... > For long messages or medium-size messages and non-power-of-two > no. of processes, we use a ring algorithm. > ..." > > Could someone please tell me what algorithm is used when we have "long messages and power-of-two no. of processes"? Thanks a lot! > > Best, > Jen > _______________________________________________ > 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

allgatherv
by Jen B 09 May '16

09 May '16

Hello, I am looking at allgatherv.c from: git.mpich.org/mpich.git/tree/HEAD:/src/mpi/coll It says the following: "For short or medium-size messages and power-of-two no. of processes, we use the recursive doubling algorithm. ... For long messages or medium-size messages and non-power-of-two no. of processes, we use a ring algorithm. ..." Could someone please tell me what algorithm is used when we have "long messages and power-of-two no. of processes"? Thanks a lot! Best, Jen _______________________________________________ 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] MPI_Comm_accept() pins cpu core 100%
by Dmitriy Lyubimov 09 May '16

09 May '16

ok, thank you. What implications does this have? does this reduce the solution capabilities/performance in any specific way? for now, i ended up just bumping process priority to the max priority for the duration of accept call, which is still not great but i don't want to mess with the build capabilities or ask people to install a custom build. On Mon, May 9, 2016 at 6:35 AM, Balaji, Pavan <balaji(a)anl.gov> wrote: > > For now, the only workaround for this is to configure with > --with-device=ch3:sock > > -- Pavan > > > On May 6, 2016, at 5:49 PM, Dmitriy Lyubimov <dlieu.7(a)gmail.com> wrote: > > > > Hi, > > > > I have a situation when a process my start up and execute > MPI_Comm_accept(), while waiting for others to join, which may take some > time. > > > > I have observed that all processes that wait in MPI_Comm_accept(), pin 1 > cpu core exactly 100%. > > > > Which is unfortunate in my situation since if i launch enough processes, > i pin the entire machine node very quickly -- and it does do other things > that run into a crawl. > > > > Is this a known problem? is there a workaround for this? > > > > I tried on ubuntu 14-04 LTS which installs mpich 3.0.4 by default but i > also tested 3.2 Nov/2015 and they both exhibit this very behavior. > > > > Thank you very much in advance. > > -Dmitriy > > _______________________________________________ > > 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