[mvapich-discuss] mvapich and gromacs4.6

Mon Nov 18 09:25:27 EST 2013

Dear Yang,

To add to what Jiri mentioned, if you are looking to enable GPU-aware MPI
features in MVAPICH2 please refer to the following sections of our user
guide:

Building
http://mvapich.cse.ohio-state.edu/support/user_guide_mvapich2-2.0b.html#x1-120004.5

Running
http://mvapich.cse.ohio-state.edu/support/user_guide_mvapich2-2.0b.html#x1-840006.20

But if you are looking for information on how the application uses the
GPUs, GROMACS mailing list would be a better place.

Sreeram Potluri

On Mon, Nov 18, 2013 at 7:04 AM, Jiri Kraus <jkraus at nvidia.com> wrote:

>  Dear Yang,
>
> the difference between mpirun with the machine file and mpirun_rsh is that
> the machine file contains each node only once and you are only starting 2
> processes if you change your machine file to node2, node2, node3, node3 and
> start with
>
> mpirun -machinefile nodes -np 4 mdrun_mpi
>
> it should behave like
>
>
> mpi_rsh -np 4 node2 node2 node3 node3
>
> Regarding the environment variable you should specify for GROMACS: I am
> sorry but I do not get your question. I general I would say that the
> GROMACS mailing lists can help you better with stating GROMACS than this
> list.
>
> Hope this helps
>
> Jiri
>
> Sent from my Nexus 7, I apologize for spelling errors and auto correction
> typos.
>
> -----Original Message-----
> *From:* hpc at lzu.edu.cn [hpc at lzu.edu.cn]
> *Received:* Montag, 18 Nov. 2013, 1:05
> *To:* Jiri Kraus [jkraus at nvidia.com]
> *CC:* sreeram potluri [potluri.2 at osu.edu]; mvapich-discuss [
> mvapich-discuss at cse.ohio-state.edu]
> *Subject:* Re: AW: [mvapich-discuss] mvapich and gromacs4.6
>
>  Dear  Jiri
>
>      My error caused I use the mpirun -machinefile nodes -np 2
> mdrun_mpi,nodes file is node2,node3, I change it with mpi_rsh -np 4 node2
> node2 node3 node3 mdrun_mpi,now it's work. I do not know what difference
> between the mpirun and mpi_rsh.
>
>     I have a question.when I  compile the mvapich with infiniband, I need
> specify the parameters? my infiniband is defacult install. if mvapich is
> install finished. what environment variable should specify  for gromacs?
> Thanks!
>
>
>  -----原始邮件-----
> *发件人:* "Jiri Kraus" <jkraus at nvidia.com>
> *发送时间:* 2013-11-18 12:01:16 (星期一)
> *收件人:* "hpc at lzu.edu.cn" <hpc at lzu.edu.cn>
> *抄送:* "sreeram potluri" <potluri.2 at osu.edu>, mvapich-discuss <
> mvapich-discuss at cse.ohio-state.edu>
>
> *主题:* AW: [mvapich-discuss] mvapich and gromacs4.6
>
>  Dear Yang,
>
>  the error message
>
>  "Incorrect launch configuration: mismatching number of PP MPI processes
> and GPUs per node.
> mdrun_mpi was started with 1 PP MPI process per node, but you provided 2
> GPUs."
>
> indicates that you are requesting GROMACS to use the two GPUs in the node
> but only start one MPI process per node. If you use the mpirun_rsh launcher
> you can do that by specifying each node that should be used twice. E.g. if
> you want to run with 4 processes on 2 nodes which are named node0 and node1
> do
>
>  mpirun_rsh -np 4 node0 node0 node1 node1 mdrun_mpi ...
>
>  If you use another launcher and depending on how your MVAPICH build is
> integrated with a possibly running batch system the procedure might be
> different.
>
>  If that does not fix the issue please send the complete commandline with
> which you try to start GROMACS, e.g.
>
>  mpirun_rsh -np 8 mdrun_mpi -v
>
>  Jiri
>
>  Sent from my smartphone, I apologize for spelling errors and auto
> correction typos.
>
> ----- Reply message -----
> Von: "hpc at lzu.edu.cn" <hpc at lzu.edu.cn>
> An: "Jiri Kraus" <jkraus at nvidia.com>
> Cc: "sreeram potluri" <potluri.2 at osu.edu>, "mvapich-discuss" <
> mvapich-discuss at cse.ohio-state.edu>
> Betreff: [mvapich-discuss] mvapich and gromacs4.6
> Datum: So., Nov 17, 2013 18:24
>
>  Dear Jiri
>
>     Thanks! the output error is:
>
> Error on node 0, will try to stop all the nodes
> Halting parallel program mdrun_mpi on CPU 0 out of 2
>
> gcq#319: "Your Country Raised You, Your Country Fed You, and Just Like Any
> Other Country it Will Break You" (Gogol Bordello)
>
> [cli_0]: aborting job:
> application called MPI_Abort(MPI_COMM_WORLD, -1) - process 0
>
> -------------------------------------------------------
> Program mdrun_mpi, VERSION 4.6.3
> Source code file: /home/*/gromacs-4.6.3/src/gmxlib/gmx_detect_hardware.c,
> line: 349
>
> Fatal error:
> Incorrect launch configuration: mismatching number of PP MPI processes and
> GPUs per node.
> mdrun_mpi was started with 1 PP MPI process per node, but you provided 2
> GPUs.
> For more information and tips for troubleshooting, please check the GROMACS
> website at http://www.gromacs.org/Documentation/Errors
> -------------------------------------------------------
>
> "We Can Dance Like Iggy Pop" (Red Hot Chili Peppers)
>
> Error on node 1, will try to stop all the nodes
> Halting parallel program mdrun_mpi on CPU 1 out of 2
>
> gcq#5: "We Can Dance Like Iggy Pop" (Red Hot Chili Peppers)
>
> [cli_1]: aborting job:
> application called MPI_Abort(MPI_COMM_WORLD, -1) - process 1
>
>
>
>
> -----原始邮件-----
> *发件人:* "Jiri Kraus" <jkraus at nvidia.com>
> *发送时间:* 2013-11-17 11:20:48 (星期日)
> *收件人:* "sreeram potluri" <potluri.2 at osu.edu>, "hpc at lzu.edu.cn" <
> hpc at lzu.edu.cn>
> *抄送:* mvapich-discuss <mvapich-discuss at cse.ohio-state.edu>
> *主题:* AW: [mvapich-discuss] mvapich and gromacs4.6
>
>  Dear Yang,
>
>  please also add the output of mdrun. GROMACS supports one GPU per MPI
> rank so you need to start as many ranks per node as there are GPUs in a
> node to utilize all GPUs.
>
>  Hope this helps
>
>  Jiri
>
>  Sent from my smartphone, I apologize for spelling errors and auto
> correction typos.
>
> ----- Reply message -----
> Von: "sreeram potluri" <potluri.2 at osu.edu>
> An: "hpc at lzu.edu.cn" <hpc at lzu.edu.cn>
> Cc: "mvapich-discuss" <mvapich-discuss at cse.ohio-state.edu>
> Betreff: [mvapich-discuss] mvapich and gromacs4.6
> Datum: Sa., Nov 16, 2013 18:11
>
>  Dear Yang,
>
>  To make sure I understand you correctly, you are seeing an issue when
> running Gromacs across two or more GPUs on the same node, using MVAPICH2?
>
>  MVAPICH2 does has support for GPU-GPU MPI communication on multi-GPU
> nodes. Can you give some more information below about the issue?
>
>  the configure options you used to build the library
>
>  the command you are using for the run
>
>  a backtrace, if its a hang, or the error, if its a failure
>
>  configuration of the nodes
>
>  This will help us see what could be going on.
>
>  Best
> Sreeram Potluri
>
>
> On Fri, Nov 15, 2013 at 7:28 AM, <hpc at lzu.edu.cn> wrote:
>
>> Dear All
>>
>>    I want to install gromacs4.6 at infiniband network.I installed the
>> mvapich2-2.0a, gromacs4.6 support the GPUs,but when I running mpirun with
>> mvapich,I found mvapich is only running one GPU, if I use two or more GPU,
>> gromacs is not running. Anyone meet this problem? openmpi is support
>> muliti-GPU.I don not know if the mvapich is only support one GPU. Maybe I
>> install process have problem. anyone have suggestion for me, I am very
>> appreciate. thanks!
>>
>> Best wishes
>>
>>
>> yang
>>
>> _______________________________________________
>> mvapich-discuss mailing list
>> mvapich-discuss at cse.ohio-state.edu
>> http://mailman.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>>
>>
>>
>
> -----------------------------------------------------------------------------------
> Nvidia GmbH
> Würselen
>
> Amtsgericht Aachen
> HRB 8361
> Managing Director: Karen Theresa Burns
>
>
> -----------------------------------------------------------------------------------
> This email message is for the sole use of the intended recipient(s) and
> may contain
> confidential information.  Any unauthorized review, use, disclosure or
> distribution
> is prohibited.  If you are not the intended recipient, please contact the
> sender by
> reply email and destroy all copies of the original message.
>
> -----------------------------------------------------------------------------------
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20131118/fe80526b/attachment-0001.html>