- discuss - mailman3.cels.anl.gov

Re: [mpich-discuss] MPI_THREAD_MULTIPLE
by Jeff Hammond 01 Jul '13

01 Jul '13

I have done my own studies of this but only in detail on Blue Gene/Q, which supports both fine-grain and fat locking to support MPI_THREAD_MULTIPLE. I believe that CrayMPI takes a noticeable hit in MPI_THREAD_MULTIPLE because their network hardware is very low latency and the software overhead associated with locking (which is fat in their case AFAIK) is noticeable by comparison. Is that the vendor in question? Do you have the option to aggregate communication and or otherwise use MPI_THREAD_SERIALIZED instead? If not, then there really isn't an alternative so comparative study will only make you a sad panda. However, if you can use MPI_THREAD_SERIALIZED, perhaps with some overhead, then you can compare the two implementations. It would be helpful if you could share code and system details. Best, Jeff On Mon, Jul 1, 2013 at 2:05 AM, Bobby Philip <bphilip.kondekeril(a)gmail.com> wrote: > Hi: > > Are there any studies that have been done on the effect of turning on > MPI_THREAD_MULTIPLE with the latest versions of MPICH? I have an AMR > application where halo or ghost updates require lots of small messages to be > sent/rec'd and I am currently seeing a performance hit with a vendor > specific implementation based on MPICH2 and am trying to see whether there > are any implementations out there that might deliver better performance. > > Thanks, > Bobby > > _______________________________________________ > 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

Re: [mpich-discuss] blcr problem
by Wesley Bland 01 Jul '13

01 Jul '13

There's been lots of turnover on the project recently so the previous experts for checkpoint/restart in MPICH are gone, but from what I can tell, it appears that it might currently be broken. This ticket (https://trac.mpich.org/projects/mpich/ticket/1144) is the only documentation about the subject that I can find, but I think the assumption is that everything has bit rotted and will need to be updated before it works again. Wesley On Jul 1, 2013, at 10:19 AM, basma a.azeem <basmaabdelazeem(a)hotmail.com> wrote: > i created a checkpoint file of the integer sort NPB after 60 sec of running the application , the checkpoint file size is 121.6 MB (121,633,453 bytes). > > i have Blcr for checkpoint/restart > when i try to restart from the checkpoint file i had , nothing happened , it just hangs > what i did wrong ? > this is the command i used. > > > basma@basma-Satellite-A500:~$ mpiexec -ckpointlib blcr \ > > -ckpoint-prefix /home/basma/ckpts/app.ckpoint \ > > -ckpoint-num 0 -n 4 > > _______________________________________________ > 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] execvp error on file comm2_spawn_child
by Anne-Sophie Mouronval 01 Jul '13

01 Jul '13

Wesley, Thanks a lot for your help. I couldn't fix the bug using the key "wdir" of a MPI_INFO object, I had to use the key "path" to fix it. Now everything is OK. Regards, Anne-Sophie Le 7/1/13 4:23 PM, Wesley Bland a ï¿½crit : > The correct way of calling MPI_COMM_SPAWN is to create an MPI_INFO object with they key wdir set to the working directory you want to use for the spawned processes. The working directory for the parent process is not necessarily used for the spawned children. Alternatively, you can just use the full path and you don't need to worry about the working directory unless you have other parts of your application which use it (reading files, etc.). > > Wesley > > P.S. We no longer call the project MPICH2. It's just MPICH now as of version 3.x > > On Jul 1, 2013, at 8:19 AM, Anne-Sophie Mouronval <anne-sophie.mouronval(a)ecp.fr> wrote: > >> Hi, >> >> I've just installed Mpich2 (using 3.0.4 sources) on my computer and every thing works fine. >> >> Now, I have to work on MPI_Comm_spawn and I have some difficulties. >> >> I've found a nice example that is provided with Mpich2 sources in >> Mpich2/share/examples/logging >> This example is based on 2 files : comm2_spawn_parent.c and comm2_spawn_child.c. >> >> I've compiled these two files but I have an error when I try to run this test : >> >> $ mpiexec -np 2 ./comm2_spawn_parent >> [proxy:1:0@bambou] HYDU_create_process (./utils/launch/launch.c:75): execvp error on file comm2_spawn_child (No such file or directory) >> >> However, "comm2_spawn_child" is present in the current directory. >> >> Thus I've tried to use a "-path" option but it's not recognized : >> $mpiexec -path mypath -np 2 ./comm2_spawn_parent >> match_arg (./utils/args/args.c:160): unrecognized argument path >> >> I've looked for some documentation of this test but I haven't found anything. >> >> Regards, >> Anne-Sophie >> _______________________________________________ >> discuss mailing list discuss(a)mpich.org >> To manage subscription options or unsubscribe: >> https://lists.mpich.org/mailman/listinfo/discuss >

1 0

blcr problem
by basma a.azeem 01 Jul '13

01 Jul '13

i created a checkpoint file of the integer sort NPB after 60 sec of running the application , the checkpoint file size is 121.6 MB (121,633,453 bytes). i have Blcr for checkpoint/restart when i try to restart from the checkpoint file i had , nothing happened , it just hangs what i did wrong ? this is the command i used. basma@basma-Satellite-A500:~$ mpiexec -ckpointlib blcr \ > -ckpoint-prefix /home/basma/ckpts/app.ckpoint \ > -ckpoint-num 0 -n 4

1 0

Re: [mpich-discuss] execvp error on file comm2_spawn_child
by Wesley Bland 01 Jul '13

01 Jul '13

The correct way of calling MPI_COMM_SPAWN is to create an MPI_INFO object with they key wdir set to the working directory you want to use for the spawned processes. The working directory for the parent process is not necessarily used for the spawned children. Alternatively, you can just use the full path and you don't need to worry about the working directory unless you have other parts of your application which use it (reading files, etc.). Wesley P.S. We no longer call the project MPICH2. It's just MPICH now as of version 3.x On Jul 1, 2013, at 8:19 AM, Anne-Sophie Mouronval <anne-sophie.mouronval(a)ecp.fr> wrote: > Hi, > > I've just installed Mpich2 (using 3.0.4 sources) on my computer and every thing works fine. > > Now, I have to work on MPI_Comm_spawn and I have some difficulties. > > I've found a nice example that is provided with Mpich2 sources in > Mpich2/share/examples/logging > This example is based on 2 files : comm2_spawn_parent.c and comm2_spawn_child.c. > > I've compiled these two files but I have an error when I try to run this test : > > $ mpiexec -np 2 ./comm2_spawn_parent > [proxy:1:0@bambou] HYDU_create_process (./utils/launch/launch.c:75): execvp error on file comm2_spawn_child (No such file or directory) > > However, "comm2_spawn_child" is present in the current directory. > > Thus I've tried to use a "-path" option but it's not recognized : > $mpiexec -path mypath -np 2 ./comm2_spawn_parent > match_arg (./utils/args/args.c:160): unrecognized argument path > > I've looked for some documentation of this test but I haven't found anything. > > Regards, > Anne-Sophie > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss

1 0

execvp error on file comm2_spawn_child
by Anne-Sophie Mouronval 01 Jul '13

01 Jul '13

Hi, I've just installed Mpich2 (using 3.0.4 sources) on my computer and every thing works fine. Now, I have to work on MPI_Comm_spawn and I have some difficulties. I've found a nice example that is provided with Mpich2 sources in Mpich2/share/examples/logging This example is based on 2 files : comm2_spawn_parent.c and comm2_spawn_child.c. I've compiled these two files but I have an error when I try to run this test : $ mpiexec -np 2 ./comm2_spawn_parent [proxy:1:0@bambou] HYDU_create_process (./utils/launch/launch.c:75): execvp error on file comm2_spawn_child (No such file or directory) However, "comm2_spawn_child" is present in the current directory. Thus I've tried to use a "-path" option but it's not recognized : $mpiexec -path mypath -np 2 ./comm2_spawn_parent match_arg (./utils/args/args.c:160): unrecognized argument path I've looked for some documentation of this test but I haven't found anything. Regards, Anne-Sophie

1 0

Re: [mpich-discuss] mpich2 and windows server 2012
by Lars Lindbom 01 Jul '13

01 Jul '13

Hello, Thanks for the tips. I had tried the first without success last week (-n 1 nonmem.exe). The second with the absolute path to the example failed with the same error. Credentials for Lars rejected connecting to hostname Aborting: Unable to connect to hostname This sounds like a simple switch to allow MPI connections but I fail to identify what to change. Lars 2013/6/28 Jayesh Krishna <jayesh(a)mcs.anl.gov> > Hi, > Did you try, > > "c:\Program Files (x86)\MPICH2\bin\mpiexec.exe" -n 1 nonmem.exe > > "c:\Program Files (x86)\MPICH2\bin\mpiexec.exe" -n 2 "c:\Program Files > (x86)\MPICH2\examples\cpi.exe" > > Regards, > Jayesh > > ----- Original Message ----- > From: "Costas Yamin" <costas.yamin(a)gmail.com> > To: discuss(a)mpich.org > Sent: Thursday, June 27, 2013 10:16:02 AM > Subject: Re: [mpich-discuss] mpich2 and windows server 2012 > > > > Hi, > > I have installed mpich2 on a Windows 2012 Essentials server machine and I > have successfully sent jobs to it remotely without doing anything different > than in my own workstation (Windows8 x64). I have configured mpiexec to use > particular ports and added the relevant rule in windows firewall. So there > shouldn't be any compatibility issues arising from the OS itself... I > assume you installed mpich2 as an Administrator. > > Costas > > > On 27/6/2013 10:59 πμ, Lars Lindbom wrote: > > > > > > Hi, > > > > I have a problem getting mpich2 to run on Windows Server 2012. We have a > small set of servers successfully configured and running Windows Server > 2008 R2 and MPICH2 without any problem. > > > > The error I get seems to indicate a problem in the authentication process. > > > > >"c:\Program Files (x86)\MPICH2\bin\mpiexec.exe" –validate > > SUCCESS > > > > >"c:\Program Files (x86)\MPICH2\bin\mpiexec.exe" -host localhost nonmem.exe > > Credentials for Lars rejected connecting to localhost > > Aborting: Unable to connect to localhost > > > > The credentials are correct and I have tried multiple user accounts with > the same result. > > I don't think it's related but for what it's worth I have made sure that I > have the same firewall settings for the mpich executables as on the 2008 R2 > servers. > > > > > I would appreciate any help in getting this solved. > > > > > Thanks, > > > > > Lars > > > > > _______________________________________________ > 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

MPI_THREAD_MULTIPLE
by Bobby Philip 01 Jul '13

01 Jul '13

Hi: Are there any studies that have been done on the effect of turning on MPI_THREAD_MULTIPLE with the latest versions of MPICH? I have an AMR application where halo or ghost updates require lots of small messages to be sent/rec'd and I am currently seeing a performance hit with a vendor specific implementation based on MPICH2 and am trying to see whether there are any implementations out there that might deliver better performance. Thanks, Bobby

1 0

Re: [mpich-discuss] Non-blocking collectives
by rreddypsc＠gmail.com 29 Jun '13

29 Jun '13

This is a very good example! Thanks for the clarification! Thanks, Raghu --On Saturday, June 29, 2013 3:08 PM -0500 Rajeev Thakur <thakur(a)mcs.anl.gov> wrote: > Yes. > > On Jun 29, 2013, at 3:05 PM, Jed Brown wrote: > >> Rajeev Thakur <thakur(a)mcs.anl.gov> writes: >> >>> It doesn't have to be a matching Wait. The implementation is required >>> to make progress on other pending communication while it is blocked in >>> a Wait. >> >> Or blocked in any of a number of other places, if non-blocking >> collective requests have the same completion semantics as point-to-point >> requests. E.g., this shouldn't deadlock: >> >> MPI_Iallreduce(...,COMM_WORLD,&req); >> if (!rank) MPI_Wait(&req); >> MPI_Allgather(...,COMM_WORLD); >> if (rank) MPI_Wait(&req); > > _______________________________________________ > 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] Non-blocking collectives
by Rajeev Thakur 29 Jun '13

29 Jun '13

Yes. On Jun 29, 2013, at 3:05 PM, Jed Brown wrote: > Rajeev Thakur <thakur(a)mcs.anl.gov> writes: > >> It doesn't have to be a matching Wait. The implementation is required >> to make progress on other pending communication while it is blocked in >> a Wait. > > Or blocked in any of a number of other places, if non-blocking > collective requests have the same completion semantics as point-to-point > requests. E.g., this shouldn't deadlock: > > MPI_Iallreduce(...,COMM_WORLD,&req); > if (!rank) MPI_Wait(&req); > MPI_Allgather(...,COMM_WORLD); > if (rank) MPI_Wait(&req);

1 0