Re: [gmx-users] Multithread run issues
Hi, That user's input can't use OpenMP, like the message says. So try the other options. (And again, don't run mpirun on a thread-MPI mdrun!) Mark On Mon, Oct 10, 2016 at 1:02 PM Mahmood Naderanwrote: > Sorry for the previous incomplete email. > > Program mdrun, VERSION 5.1 > Source code file: > /share/apps/chemistry/gromacs-5.1/src/programs/mdrun/resource-division.cpp, > line: 746 > > Fatal error: > OpenMP threads have been requested with cut-off scheme Group, but these are > only supported with cut-off scheme Verlet > For more information and tips for troubleshooting, please check the GROMACS > website at http://www.gromacs.org/Documentation/Errors > > > > I read that document from that web site but didn't understand what is the > issue! > Thanks > > Regards, > Mahmood > > > > On Mon, Oct 10, 2016 at 2:29 PM, Mahmood Naderan > wrote: > > > >mpirun -np 1 mdrun_mpi -v -ntomp 2 > > > > Agree with that. > > > > >This is not a problem to solve by running applications differently. You > > >will have users running jobs with single ranks/processes that use > > threading > > >of various kinds to fill the cores. That's a feature, not a bug. Either > > >configure PBS to cope with decades-old technology, or don't worry about > > it. > > > > I found this document > > ( > > https://wiki.anl.gov/cnm/HPC/Submitting_and_Managing_Jobs/ > > Advanced_node_selection#Multithreading_using_OpenMP) > > That is what I want to be sure that number of threads and cores used for > > gromacs fits to the PBS stats. > > > > Instead of the variable, I wrote > > > > > > > > #PBS -l nodes=1:ppn=2 > > export OMP_NUM_THREADS=2 > > mpirun mdrun -v > > > > So that will use two cores with 4 threads totally and PBS should report 4 > > processors are occupied. > > However, gromacs failed with the following error > > > > > > > > > > Regards, > > Mahmood > > > > > > > > On Mon, Oct 10, 2016 at 1:41 PM, Mark Abraham > > wrote: > > > >> Hi, > >> > >> Yeah, but that run is very likely > >> > >> a) useless because you're just running two copies of the same simulation > >> because you're not running MPI-enabled mdrun > >> b) and even if not, less efficient than the thread-MPI version > >> > >> mdrun -v -nt 2 > >> > >> c) and even if not, likely slightly less efficient than the real-MPI > >> version > >> > >> mpirun -np 1 mdrun_mpi -v -ntomp 2 > >> > >> top isn't necessarily reporting anything relevant. A CPU can be > nominally > >> idle while waiting for communication, but what does top think about > that? > >> > >> Mark > >> > >> On Mon, Oct 10, 2016 at 11:47 AM Mahmood Naderan > >> wrote: > >> > >> > OK. I understood the documents. > >> > Thing that I want is to see two processes (for example) each consumes > >> 100% > >> > cpu. The command for that is > >> > > >> > mpirun -np 2 mdrun -v -nt 1 > >> > > >> > Thanks Mark. > >> > > >> > Regards, > >> > Mahmood > >> > -- > >> > Gromacs Users mailing list > >> > > >> > * Please search the archive at > >> > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before > >> > posting! > >> > > >> > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists > >> > > >> > * For (un)subscribe requests visit > >> > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or > >> > send a mail to gmx-users-requ...@gromacs.org. > >> -- > >> Gromacs Users mailing list > >> > >> * Please search the archive at http://www.gromacs.org/Support > >> /Mailing_Lists/GMX-Users_List before posting! > >> > >> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists > >> > >> * For (un)subscribe requests visit > >> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or > >> send a mail to gmx-users-requ...@gromacs.org. > >> > > > > > -- > Gromacs Users mailing list > > * Please search the archive at > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before > posting! > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists > > * For (un)subscribe requests visit > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or > send a mail to gmx-users-requ...@gromacs.org. -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Multithread run issues
Sorry for the previous incomplete email. Program mdrun, VERSION 5.1 Source code file: /share/apps/chemistry/gromacs-5.1/src/programs/mdrun/resource-division.cpp, line: 746 Fatal error: OpenMP threads have been requested with cut-off scheme Group, but these are only supported with cut-off scheme Verlet For more information and tips for troubleshooting, please check the GROMACS website at http://www.gromacs.org/Documentation/Errors I read that document from that web site but didn't understand what is the issue! Thanks Regards, Mahmood On Mon, Oct 10, 2016 at 2:29 PM, Mahmood Naderanwrote: > >mpirun -np 1 mdrun_mpi -v -ntomp 2 > > Agree with that. > > >This is not a problem to solve by running applications differently. You > >will have users running jobs with single ranks/processes that use > threading > >of various kinds to fill the cores. That's a feature, not a bug. Either > >configure PBS to cope with decades-old technology, or don't worry about > it. > > I found this document > ( > https://wiki.anl.gov/cnm/HPC/Submitting_and_Managing_Jobs/ > Advanced_node_selection#Multithreading_using_OpenMP) > That is what I want to be sure that number of threads and cores used for > gromacs fits to the PBS stats. > > Instead of the variable, I wrote > > > > #PBS -l nodes=1:ppn=2 > export OMP_NUM_THREADS=2 > mpirun mdrun -v > > So that will use two cores with 4 threads totally and PBS should report 4 > processors are occupied. > However, gromacs failed with the following error > > > > > Regards, > Mahmood > > > > On Mon, Oct 10, 2016 at 1:41 PM, Mark Abraham > wrote: > >> Hi, >> >> Yeah, but that run is very likely >> >> a) useless because you're just running two copies of the same simulation >> because you're not running MPI-enabled mdrun >> b) and even if not, less efficient than the thread-MPI version >> >> mdrun -v -nt 2 >> >> c) and even if not, likely slightly less efficient than the real-MPI >> version >> >> mpirun -np 1 mdrun_mpi -v -ntomp 2 >> >> top isn't necessarily reporting anything relevant. A CPU can be nominally >> idle while waiting for communication, but what does top think about that? >> >> Mark >> >> On Mon, Oct 10, 2016 at 11:47 AM Mahmood Naderan >> wrote: >> >> > OK. I understood the documents. >> > Thing that I want is to see two processes (for example) each consumes >> 100% >> > cpu. The command for that is >> > >> > mpirun -np 2 mdrun -v -nt 1 >> > >> > Thanks Mark. >> > >> > Regards, >> > Mahmood >> > -- >> > Gromacs Users mailing list >> > >> > * Please search the archive at >> > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before >> > posting! >> > >> > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists >> > >> > * For (un)subscribe requests visit >> > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or >> > send a mail to gmx-users-requ...@gromacs.org. >> -- >> Gromacs Users mailing list >> >> * Please search the archive at http://www.gromacs.org/Support >> /Mailing_Lists/GMX-Users_List before posting! >> >> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists >> >> * For (un)subscribe requests visit >> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or >> send a mail to gmx-users-requ...@gromacs.org. >> > > -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Multithread run issues
>mpirun -np 1 mdrun_mpi -v -ntomp 2 Agree with that. >This is not a problem to solve by running applications differently. You >will have users running jobs with single ranks/processes that use threading >of various kinds to fill the cores. That's a feature, not a bug. Either >configure PBS to cope with decades-old technology, or don't worry about it. I found this document ( https://wiki.anl.gov/cnm/HPC/Submitting_and_Managing_Jobs/Advanced_node_selection#Multithreading_using_OpenMP ) That is what I want to be sure that number of threads and cores used for gromacs fits to the PBS stats. Instead of the variable, I wrote #PBS -l nodes=1:ppn=2 export OMP_NUM_THREADS=2 mpirun mdrun -v So that will use two cores with 4 threads totally and PBS should report 4 processors are occupied. However, gromacs failed with the following error Regards, Mahmood On Mon, Oct 10, 2016 at 1:41 PM, Mark Abrahamwrote: > Hi, > > Yeah, but that run is very likely > > a) useless because you're just running two copies of the same simulation > because you're not running MPI-enabled mdrun > b) and even if not, less efficient than the thread-MPI version > > mdrun -v -nt 2 > > c) and even if not, likely slightly less efficient than the real-MPI > version > > mpirun -np 1 mdrun_mpi -v -ntomp 2 > > top isn't necessarily reporting anything relevant. A CPU can be nominally > idle while waiting for communication, but what does top think about that? > > Mark > > On Mon, Oct 10, 2016 at 11:47 AM Mahmood Naderan > wrote: > > > OK. I understood the documents. > > Thing that I want is to see two processes (for example) each consumes > 100% > > cpu. The command for that is > > > > mpirun -np 2 mdrun -v -nt 1 > > > > Thanks Mark. > > > > Regards, > > Mahmood > > -- > > Gromacs Users mailing list > > > > * Please search the archive at > > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before > > posting! > > > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists > > > > * For (un)subscribe requests visit > > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or > > send a mail to gmx-users-requ...@gromacs.org. > -- > Gromacs Users mailing list > > * Please search the archive at http://www.gromacs.org/ > Support/Mailing_Lists/GMX-Users_List before posting! > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists > > * For (un)subscribe requests visit > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or > send a mail to gmx-users-requ...@gromacs.org. > -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Multithread run issues
Hi, Yeah, but that run is very likely a) useless because you're just running two copies of the same simulation because you're not running MPI-enabled mdrun b) and even if not, less efficient than the thread-MPI version mdrun -v -nt 2 c) and even if not, likely slightly less efficient than the real-MPI version mpirun -np 1 mdrun_mpi -v -ntomp 2 top isn't necessarily reporting anything relevant. A CPU can be nominally idle while waiting for communication, but what does top think about that? Mark On Mon, Oct 10, 2016 at 11:47 AM Mahmood Naderanwrote: > OK. I understood the documents. > Thing that I want is to see two processes (for example) each consumes 100% > cpu. The command for that is > > mpirun -np 2 mdrun -v -nt 1 > > Thanks Mark. > > Regards, > Mahmood > -- > Gromacs Users mailing list > > * Please search the archive at > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before > posting! > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists > > * For (un)subscribe requests visit > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or > send a mail to gmx-users-requ...@gromacs.org. -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Multithread run issues
OK. I understood the documents. Thing that I want is to see two processes (for example) each consumes 100% cpu. The command for that is mpirun -np 2 mdrun -v -nt 1 Thanks Mark. Regards, Mahmood -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Multithread run issues
Hi, On Sun, Oct 9, 2016 at 2:06 PM Mahmood Naderanwrote: > Hi, > > Users issue the command "mdrun -v" and that will automatically read input > > files in the working directory. There are two issue with that which I am > > not aware of the solution. > > > > 1- How the number of cores can be changed? > http://manual.gromacs.org/documentation/2016/user-guide/mdrun-performance.html > > 2- Viewing the output of "top" command, it is saying that mdrun uses 400% > > cpu. That mean 4 cores are occupied. Problem is that we prefer to see four > > processes each consumes 100% cpu. Why? Because in the first situation, PBS > > wrongly sees that one cores is used (while 4 cores are used), but in the > > later, it will correctly sees 4 cores. > This is not a problem to solve by running applications differently. You will have users running jobs with single ranks/processes that use threading of various kinds to fill the cores. That's a feature, not a bug. Either configure PBS to cope with decades-old technology, or don't worry about it. :-) As I read the documentation, I think the replacement of "mdrun -v" should be > > > > mpirun -np N mdrun -v > > > > Am I correct? > No, because that mdrun isn't configured with MPI support (else it would be called mdrun_mpi). If it was, then mpirun -np N mdrun_mpi is plausible, but less efficient than the alternatives. Your mdrun is using the built in thread-MPI that you can read about at that link. Mark > > > > Regards, > > Mahmood > > -- > > Gromacs Users mailing list > > > > * Please search the archive at > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before > posting! > > > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists > > > > * For (un)subscribe requests visit > > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or > send a mail to gmx-users-requ...@gromacs.org. > > -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
[gmx-users] Multithread run issues
Hi, Users issue the command "mdrun -v" and that will automatically read input files in the working directory. There are two issue with that which I am not aware of the solution. 1- How the number of cores can be changed? 2- Viewing the output of "top" command, it is saying that mdrun uses 400% cpu. That mean 4 cores are occupied. Problem is that we prefer to see four processes each consumes 100% cpu. Why? Because in the first situation, PBS wrongly sees that one cores is used (while 4 cores are used), but in the later, it will correctly sees 4 cores. As I read the documentation, I think the replacement of "mdrun -v" should be mpirun -np N mdrun -v Am I correct? Regards, Mahmood -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.