This seems to be a problem with your MPI library. Test to see whether other MPI programs don't have the same problem. If it is not GROMACS specific please ask on the mailinglist of your MPI library. If it only happens with GROMACS be more specific about what your setup is (what MPI library, what hardware, ...).
Also you could use the latest GROMACS 4.5.x. It has built in thread support and doesn't need MPI as long as you only run on n cores within one SMP node. Roland On Wed, Apr 27, 2011 at 2:13 PM, Hrachya Astsatryan <hr...@sci.am> wrote: > Dear Mark Abraham & all, > > We used another benchmarking systems, such as d.dppc on 4 processors, but > we have the same problem (1 proc use about 100%, the others 0%). > After for a while we receive the following error: > > Working directory is /localuser/armen/d.dppc > Running on host wn1.ysu-cluster.grid.am > Time is Fri Apr 22 13:55:47 AMST 2011 > Directory is /localuser/armen/d.dppc > ____START____ > Start: Fri Apr 22 13:55:47 AMST 2011 > p2_487: p4_error: Timeout in establishing connection to remote process: 0 > rm_l_2_500: (301.160156) net_send: could not write to fd=5, errno = 32 > p2_487: (301.160156) net_send: could not write to fd=5, errno = 32 > p0_32738: p4_error: net_recv read: probable EOF on socket: 1 > p3_490: (301.160156) net_send: could not write to fd=6, errno = 104 > p3_490: p4_error: net_send write: -1 > p3_490: (305.167969) net_send: could not write to fd=5, errno = 32 > p0_32738: (305.371094) net_send: could not write to fd=4, errno = 32 > p1_483: p4_error: net_recv read: probable EOF on socket: 1 > rm_l_1_499: (305.167969) net_send: could not write to fd=5, errno = 32 > p1_483: (311.171875) net_send: could not write to fd=5, errno = 32 > Fri Apr 22 14:00:59 AMST 2011 > End: Fri Apr 22 14:00:59 AMST 2011 > ____END____ > > We tried new version of Gromacs, but receive the same error. > Please, help us to overcome the problem. > > > With regards, > Hrach > > > On 4/22/11 1:41 PM, Mark Abraham wrote: > >> On 4/22/2011 5:40 PM, Hrachya Astsatryan wrote: >> >>> Dear all, >>> >>> I would like to inform you that I have installed the gromacs4.0.7 package >>> on the cluster (nodes of the cluster are 8 core Intel, OS: RHEL4 Scientific >>> Linux) with the following steps: >>> >>> yum install fftw3 fftw3-devel >>> ./configure --prefix=/localuser/armen/gromacs --enable-mpi >>> >>> Also I have downloaded gmxbench-3.0 package and try to run d.villin to >>> test it. >>> >>> Unfortunately it wok fine until np is 1,2,3, if I use more than 3 procs I >>> receive low CPU balancing and the process in hanging. >>> >>> Could you, please, help me to overcome the problem? >>> >> >> Probably you have only four physical cores (hyperthreading is not normally >> useful), or your MPI is configured to use only four cores, or these >> benchmarks are too small to scale usefully. >> >> Choosing to do a new installation of a GROMACS version that is several >> years old is normally less productive than the latest version. >> >> Mark >> >> >> >> > -- > gmx-users mailing list gmx-users@gromacs.org > http://lists.gromacs.org/mailman/listinfo/gmx-users > Please search the archive at > http://www.gromacs.org/Support/Mailing_Lists/Search before posting! > Please don't post (un)subscribe requests to the list. Use the www interface > or send it to gmx-users-requ...@gromacs.org. > Can't post? Read http://www.gromacs.org/Support/Mailing_Lists > > > -- ORNL/UT Center for Molecular Biophysics cmb.ornl.gov 865-241-1537, ORNL PO BOX 2008 MS6309
-- gmx-users mailing list gmx-users@gromacs.org http://lists.gromacs.org/mailman/listinfo/gmx-users Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/Search before posting! Please don't post (un)subscribe requests to the list. Use the www interface or send it to gmx-users-requ...@gromacs.org. Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
