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

Re: [mpich-discuss] Understanding recursive doubling impl. in Mpich
by Dorier, Matthieu 29 Nov '16

29 Nov '16

I see, thank you! Matthieu ________________________________________ From: Thakur, Rajeev [thakur(a)anl.gov] Sent: Tuesday, November 29, 2016 1:37 PM To: discuss(a)mpich.org Subject: Re: [mpich-discuss] Understanding recursive doubling impl. in Mpich I don’t remember for sure, but probably because that code is meant to work even for non power of two numbers of processes (although the if statement precludes that right now) , and calculating the recvcount accurately in that case is more work. Rajeev > On Nov 29, 2016, at 1:14 PM, Dorier, Matthieu <mdorier(a)anl.gov> wrote: > > Hi, > > I'm looking at Mpich's code for MPI_Allgather (src/mpi/coll/allgather.c), lines 183 to 222, where the recursive doubling algorithm is used. I''m under the impression that the MPI_Sendrecv operation used line 206 is not issued with the same recvcount value than the sendcount value of the sending process. > > Example: if I compute manually for an allgather across 4 processes: in the first loop iteration, process 0 and 1 will exchange data, and process 2 and 3 will do the same. Focusing on 0 and 1, process 0 sends 1 item and expects 3, process 1 sends 1 item and expects 4. > > I know that posting a receive with a size larger than what is sent is correct in MPI, and that would explain why line 220 we have MPIR_Get_count_impl(&status, recvtype, &last_recv_cnt); > I just want to make sure that I'm not missing something, and if not, I'd like to know why processes overestimate the received size instead of doubling the curr_cnt variable at every step? Is it to allow the next part of the code (which is between /* --BEGIN EXPERIMENTAL-- */ comments) to work? > > Thanks > > Matthieu > _______________________________________________ > 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] Understanding recursive doubling impl. in Mpich
by Thakur, Rajeev 29 Nov '16

29 Nov '16

I don’t remember for sure, but probably because that code is meant to work even for non power of two numbers of processes (although the if statement precludes that right now) , and calculating the recvcount accurately in that case is more work. Rajeev > On Nov 29, 2016, at 1:14 PM, Dorier, Matthieu <mdorier(a)anl.gov> wrote: > > Hi, > > I'm looking at Mpich's code for MPI_Allgather (src/mpi/coll/allgather.c), lines 183 to 222, where the recursive doubling algorithm is used. I''m under the impression that the MPI_Sendrecv operation used line 206 is not issued with the same recvcount value than the sendcount value of the sending process. > > Example: if I compute manually for an allgather across 4 processes: in the first loop iteration, process 0 and 1 will exchange data, and process 2 and 3 will do the same. Focusing on 0 and 1, process 0 sends 1 item and expects 3, process 1 sends 1 item and expects 4. > > I know that posting a receive with a size larger than what is sent is correct in MPI, and that would explain why line 220 we have MPIR_Get_count_impl(&status, recvtype, &last_recv_cnt); > I just want to make sure that I'm not missing something, and if not, I'd like to know why processes overestimate the received size instead of doubling the curr_cnt variable at every step? Is it to allow the next part of the code (which is between /* --BEGIN EXPERIMENTAL-- */ comments) to work? > > Thanks > > Matthieu > _______________________________________________ > 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

Understanding recursive doubling impl. in Mpich
by Dorier, Matthieu 29 Nov '16

29 Nov '16

Hi, I'm looking at Mpich's code for MPI_Allgather (src/mpi/coll/allgather.c), lines 183 to 222, where the recursive doubling algorithm is used. I''m under the impression that the MPI_Sendrecv operation used line 206 is not issued with the same recvcount value than the sendcount value of the sending process. Example: if I compute manually for an allgather across 4 processes: in the first loop iteration, process 0 and 1 will exchange data, and process 2 and 3 will do the same. Focusing on 0 and 1, process 0 sends 1 item and expects 3, process 1 sends 1 item and expects 4. I know that posting a receive with a size larger than what is sent is correct in MPI, and that would explain why line 220 we have MPIR_Get_count_impl(&status, recvtype, &last_recv_cnt); I just want to make sure that I'm not missing something, and if not, I'd like to know why processes overestimate the received size instead of doubling the curr_cnt variable at every step? Is it to allow the next part of the code (which is between /* --BEGIN EXPERIMENTAL-- */ comments) to work? Thanks Matthieu _______________________________________________ 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] Can I force MPICH to use TCP communication within a node
by Kenneth Raffenetti 22 Nov '16

22 Nov '16

You can set the environment variable MPIR_CVAR_CH3_NOLOCAL=1 to force this in 3.0.4 and newer versions. Ken On 11/21/2016 11:15 PM, Kevin Buckley wrote: > Apologies in advance, as this seems so basic that > I can't belive I can't find it in the docs, however, I > can't find it in the docs. > > I have just built an MPICH 3.0.4 and have used it to compile > the SKaMPI test suite (all on an ArchLinux platform) > > I can run the test suite with a > > /path/to/mpirun -np 2 ./skampi -i ski/skampi_pt2pt.ski > > and it works, however, it's clearly (?) using the SHM channel > > ... > [proxy:0:[email protected]] PMI response: cmd=get_result rc=0 > msg=success value=(vector,(0,1,1)) > [proxy:0:[email protected]] got pmi command (from 0): put > kvsname=kvs_12233_0 key=sharedFilename[0] value=/dev/shm/mpich_shar_tmpUq7v2w > [proxy:0:[email protected]] cached command: > sharedFilename[0]=/dev/shm/mpich_shar_tmpUq7v2w > ... > > > I'm normally an OpenMPI man and I'd like to try and force > MPICH to use a TCP/IP "channel" along the lines of telling > OpenMPI to use a TCP/IP "trasnport" with a > > -mca btl tcp,self > > so, what's the MPICH incantation to achieve the same thing ? > > Yours, clearly missing something obvious, > Kevin > _______________________________________________ > 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

Can I force MPICH to use TCP communication within a node
by Kevin Buckley 21 Nov '16

21 Nov '16

Apologies in advance, as this seems so basic that I can't belive I can't find it in the docs, however, I can't find it in the docs. I have just built an MPICH 3.0.4 and have used it to compile the SKaMPI test suite (all on an ArchLinux platform) I can run the test suite with a /path/to/mpirun -np 2 ./skampi -i ski/skampi_pt2pt.ski and it works, however, it's clearly (?) using the SHM channel ... [proxy:0:[email protected]] PMI response: cmd=get_result rc=0 msg=success value=(vector,(0,1,1)) [proxy:0:[email protected]] got pmi command (from 0): put kvsname=kvs_12233_0 key=sharedFilename[0] value=/dev/shm/mpich_shar_tmpUq7v2w [proxy:0:[email protected]] cached command: sharedFilename[0]=/dev/shm/mpich_shar_tmpUq7v2w ... I'm normally an OpenMPI man and I'd like to try and force MPICH to use a TCP/IP "channel" along the lines of telling OpenMPI to use a TCP/IP "trasnport" with a -mca btl tcp,self so, what's the MPICH incantation to achieve the same thing ? Yours, clearly missing something obvious, Kevin _______________________________________________ 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] installation error
by Kenneth Raffenetti 20 Nov '16

20 Nov '16

Your build appears to be triggering automake, which should not happen for tarballs. Can you make sure that the system time on your virtual machine is accurate? Bad time settings can cause this to happen. Ken On 11/17/2016 07:41 PM, Liang ZHENG wrote: > tried 3.3a2, but still have problems. > > btw, I am using ubuntu 16.04 through virtualbox in windows 7 > it is the first time for me to use a virtual machine for installing mpich. > > > > > On Thu, Nov 17, 2016 at 10:59 PM, Rob Latham <robl(a)mcs.anl.gov > <mailto:[email protected]>> wrote: > > > This has been fixed for a while in our git repository. What version > are you building? You could try the recently announced 3.3a2: > > http://www.mpich.org/2016/11/13/mpich-3-3a2-released/ > <http://www.mpich.org/2016/11/13/mpich-3-3a2-released/> > > ==rob > _______________________________________________ > discuss mailing list discuss(a)mpich.org <mailto:[email protected]> > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss > <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] installation error
by Liang ZHENG 17 Nov '16

17 Nov '16

tried 3.3a2, but still have problems. btw, I am using ubuntu 16.04 through virtualbox in windows 7 it is the first time for me to use a virtual machine for installing mpich. On Thu, Nov 17, 2016 at 10:59 PM, Rob Latham <robl(a)mcs.anl.gov> wrote: > > This has been fixed for a while in our git repository. What version are > you building? You could try the recently announced 3.3a2: > > http://www.mpich.org/2016/11/13/mpich-3-3a2-released/ > > ==rob > _______________________________________________ > 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] installation error
by Rob Latham 17 Nov '16

17 Nov '16

This has been fixed for a while in our git repository. What version are you building? You could try the recently announced 3.3a2: http://www.mpich.org/2016/11/13/mpich-3-3a2-released/ ==rob _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

installation error
by Liang ZHENG 17 Nov '16

17 Nov '16

_______________________________________________ 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] Hybrid HPC system
by Min Si 16 Nov '16

16 Nov '16

I guess you might need to put all the MPICH binaries (e.g., hydra_pmi_proxy) to the same path on each node. I have executed MPICH on Intel MIC chips from the host CPU node where OS are different. The thing I did was: 1. build MPICH for both CPU node and MIC on the CPU node (you have done this step). 2. upload the MIC binaries to the same path on MIC chip as on the CPU node For example: - on CPU node : /tmp/mpich/install/bin holds the CPU version - on MIC : /tmp/mpich/install/bin holds the MIC version 3. compile helloworld.c with the MIC version mpicc 4. execute on CPU node: mpiexe -np 2 -f <hostfile with mic hostnames>./helloworld I think you should be able to follow step 2, but since your helloworld binary is also built with different OS, you might want to put it also into the same path on two nodes similar as we do for MPICH binaries. Min On 11/16/16 8:29 AM, Kenneth Raffenetti wrote: > Have you disabled any and all firewalls on both nodes? It sounds like > they are unable to communicate in initialization. > > Ken > > On 11/16/2016 07:34 AM, Doha Ehab wrote: >> Yes, I built MPICH-3 on both systems and I tried the code on each node >> separately and it worked, I tried each node with other nodes that has >> the same operating system and it worked as well. >> When I try the code on the 2 nodes that have different operating systems >> no result or error message appear. >> >> Regards >> Doha >> >> On Mon, Nov 14, 2016 at 6:25 PM, Kenneth Raffenetti >> <raffenet(a)mcs.anl.gov <mailto:[email protected]>> wrote: >> >> It may be possible to run in such a setup, but it would not be >> recommended. Did you build MPICH on both systems you are trying to >> run on? What exactly happened when the code didn't work? >> >> Ken >> >> >> On 11/13/2016 12:36 AM, Doha Ehab wrote: >> >> Hello, >> I tried to run a parallel (Hello World) C code on a cluster >> that has 2 >> nodes, the nodes have different operating system so the code >> did not >> work and no results were printed. >> How to make such a cluster work? is there is extra steps that >> should be >> done? >> >> Regards, >> Doha >> >> >> _______________________________________________ >> discuss mailing list discuss(a)mpich.org >> <mailto:[email protected]> >> To manage subscription options or unsubscribe: >> https://lists.mpich.org/mailman/listinfo/discuss >> <https://lists.mpich.org/mailman/listinfo/discuss> >> >> _______________________________________________ >> discuss mailing list discuss(a)mpich.org >> <mailto:[email protected]> >> To manage subscription options or unsubscribe: >> https://lists.mpich.org/mailman/listinfo/discuss >> <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