November 2019 - discuss - mailman3.cels.anl.gov

Re: [mpich-discuss] UCX run-time warnings
by Sajid Ali 27 Nov '19

27 Nov '19

Hi Kenneth, I installed mpich-master with external ucx(a)1.6.0 and I don't get any of the UCX errors. I still have the slurmstepd error but that's a slurm error and probably has nothing to do with mpich. Thanks for your help! -- Sajid Ali | PhD Candidate Applied Physics Northwestern University s-sajid-ali.github.io

1 0

Re: [mpich-discuss] UCX run-time warnings
by Sajid Ali 27 Nov '19

27 Nov '19

Hi Kenneth, Thanks for the spack PR. I'll get back to you with the results with that soon. It looks like the links got scrubbed in the last email so I'll give them here. 1 -> https://github.com/s-sajid-ali/nu-quest-mpi/blob/master/new_mpi_bench/submi… 2 -> https://github.com/s-sajid-ali/nu-quest-mpi/tree/master/new_mpi_bench/mdb6i… 3 -> https://github.com/s-sajid-ali/nu-quest-mpi/blob/master/new_mpi_bench/mdb6i…

1 0

Re: [mpich-discuss] UCX run-time warnings
by Sajid Ali 27 Nov '19

27 Nov '19

Hi Kenneth, I have added a command to print the linking info for the benchmark executable for each mpi build as per [1]. The results for benchmarks built with mpich@develop with gcc(a)8.3.0 are here [2] where the mpi_config file [3] has the ldd info printed out and I’m confirming that the right mpi library is being used. I don’t see any hcoll linking and can confirm that it is picking up the right mpich version. I have also printed out the linking info for mpich itself and attach the output of the following along with this email : [sas4990@quser10 ~]$ ldd $(spack location -i mpich@develop%[email protected])/lib/libmpich.so &> mpich_ldd The only thing that I’m concerned about is that the libmpich.so has no link to libslurmfull.so (but openmpi does link to). Thank You, Sajid Ali | PhD Candidate Applied Physics Northwestern University s-sajid-ali.github.io

1 0

Re: [mpich-discuss] UCX run-time warnings
by Raffenetti, Kenneth J. 27 Nov '19

27 Nov '19

I have also opened a PR on spack to add a dependency on ucx for MPICH. Otherwise an embedded version is used, which *should* work but is not the preferred configuration. https://github.com/spack/spack/pull/13921 Ken On 11/27/19 2:08 PM, Raffenetti, Kenneth J. via discuss wrote: > Those warning messages are not coming from MPICH. Can you run ldd one of > the benchmark binaries and confirm all the expected libraries are linked > to it? Are you perhaps also linked against an hcoll library? > > Ken > > On 11/27/19 12:58 PM, Sajid Ali wrote: >> Hi Kenneth, >> >> Running the same test with mpich@develop, I still see the following >> errors in a typical outfile : >> >> |[sas4990@quser10 mdb6i63]$ cat lat_10 >> ---------------------------------------- ... >> ---------------------------------------- slurmstepd: error: mpi/pmi2: no >> value for key in req slurmstepd: error: mpi/pmi2: no value for key in >> req slurmstepd: error: mpi/pmi2: no value for key in req >> [1574880837.368994] [qnode5038:4743 :0] ucp_context.c:1131 UCX WARN UCP >> version is incompatible, required: 1.5, actual: 1.4 (release 0) >> slurmstepd: error: mpi/pmi2: no value for key in req slurmstepd: error: >> mpi/pmi2: no value for key in req [1574880837.377860] [qnode5039:4546 >> :0] ucp_context.c:1131 UCX WARN UCP version is incompatible, required: >> 1.5, actual: 1.4 (release 0) slurmstepd: error: mpi/pmi2: no value for >> key in req slurmstepd: error: mpi/pmi2: no value for key in req >> slurmstepd: error: mpi/pmi2: no value for key in req slurmstepd: error: >> mpi/pmi2: no value for key in req slurmstepd: error: mpi/pmi2: no value >> for key in req # OSU MPI_Get latency Test v5.6.1 # Window creation: >> MPI_Win_allocate # Synchronization: MPI_Win_flush # Size Latency (us) 1 >> 2.13 .... | >> >> >> Since UCX is a vendored dependency of MPICH, why is MPICH complaining >> about an incompatible UCX version ? >> >> >> Thank You, >> Sajid Ali | PhD Candidate >> Applied Physics >> Northwestern University >> s-sajid-ali.github.io <http://s-sajid-ali.github.io> >> > _______________________________________________ > 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] UCX run-time warnings
by Raffenetti, Kenneth J. 27 Nov '19

27 Nov '19

Those warning messages are not coming from MPICH. Can you run ldd one of the benchmark binaries and confirm all the expected libraries are linked to it? Are you perhaps also linked against an hcoll library? Ken On 11/27/19 12:58 PM, Sajid Ali wrote: > Hi Kenneth, > > Running the same test with mpich@develop, I still see the following > errors in a typical outfile : > > |[sas4990@quser10 mdb6i63]$ cat lat_10 > ---------------------------------------- ... > ---------------------------------------- slurmstepd: error: mpi/pmi2: no > value for key in req slurmstepd: error: mpi/pmi2: no value for key in > req slurmstepd: error: mpi/pmi2: no value for key in req > [1574880837.368994] [qnode5038:4743 :0] ucp_context.c:1131 UCX WARN UCP > version is incompatible, required: 1.5, actual: 1.4 (release 0) > slurmstepd: error: mpi/pmi2: no value for key in req slurmstepd: error: > mpi/pmi2: no value for key in req [1574880837.377860] [qnode5039:4546 > :0] ucp_context.c:1131 UCX WARN UCP version is incompatible, required: > 1.5, actual: 1.4 (release 0) slurmstepd: error: mpi/pmi2: no value for > key in req slurmstepd: error: mpi/pmi2: no value for key in req > slurmstepd: error: mpi/pmi2: no value for key in req slurmstepd: error: > mpi/pmi2: no value for key in req slurmstepd: error: mpi/pmi2: no value > for key in req # OSU MPI_Get latency Test v5.6.1 # Window creation: > MPI_Win_allocate # Synchronization: MPI_Win_flush # Size Latency (us) 1 > 2.13 .... | > > > Since UCX is a vendored dependency of MPICH, why is MPICH complaining > about an incompatible UCX version ? > > > Thank You, > Sajid Ali | PhD Candidate > Applied Physics > Northwestern University > s-sajid-ali.github.io <http://s-sajid-ali.github.io> >

1 0

Re: [mpich-discuss] Question about mpich bandwidth vs message-size plot
by Congiu, Giuseppe 27 Nov '19

27 Nov '19

Sajid please use latest mpich master branch from www.github.com/pmodels/mpich<http://www.github.com/pmodels/mpich>. This contains the bug fix that was merged with the following PR: https://github.com/pmodels/mpich/pull/3947 Giuseppe Congiu Postdoctoral Appointee MCS Division Argonne National Laboratory 9700 South Cass Ave., Lemont, IL 60439 On Nov 27, 2019, at 11:15 AM, Sajid Ali via discuss <discuss(a)mpich.org<mailto:[email protected]>> wrote: Hi Jeff/Giuseppe, I spoke to the computational specialist from research IT and he confirmed that this particular build of MPICH is indeed broken. As Giuseppe pointed out the hydra launcher was running multi threaded jobs on each node because it was unable to read the nodelist. When we passed a nodelist to mpirun and repeated the same test, the results looked more reasonable (~80us latency instead of ~0.05us due to ch3:TCP). Thanks for the help! -- Sajid Ali | PhD Candidate Applied Physics Northwestern University s-sajid-ali.github.io<http://s-sajid-ali.github.io/> _______________________________________________ discuss mailing list discuss(a)mpich.org<mailto:[email protected]> To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] UCX run-time warnings
by Sajid Ali 27 Nov '19

27 Nov '19

Hi Kenneth, Running the same test with mpich@develop, I still see the following errors in a typical outfile : [sas4990@quser10 mdb6i63]$ cat lat_10 ---------------------------------------- ... ---------------------------------------- slurmstepd: error: mpi/pmi2: no value for key in req slurmstepd: error: mpi/pmi2: no value for key in req slurmstepd: error: mpi/pmi2: no value for key in req [1574880837.368994] [qnode5038:4743 :0] ucp_context.c:1131 UCX WARN UCP version is incompatible, required: 1.5, actual: 1.4 (release 0) slurmstepd: error: mpi/pmi2: no value for key in req slurmstepd: error: mpi/pmi2: no value for key in req [1574880837.377860] [qnode5039:4546 :0] ucp_context.c:1131 UCX WARN UCP version is incompatible, required: 1.5, actual: 1.4 (release 0) slurmstepd: error: mpi/pmi2: no value for key in req slurmstepd: error: mpi/pmi2: no value for key in req slurmstepd: error: mpi/pmi2: no value for key in req slurmstepd: error: mpi/pmi2: no value for key in req slurmstepd: error: mpi/pmi2: no value for key in req # OSU MPI_Get latency Test v5.6.1 # Window creation: MPI_Win_allocate # Synchronization: MPI_Win_flush # Size Latency (us) 1 2.13 .... Since UCX is a vendored dependency of MPICH, why is MPICH complaining about an incompatible UCX version ? Thank You, Sajid Ali | PhD Candidate Applied Physics Northwestern University s-sajid-ali.github.io

1 0

Re: [mpich-discuss] Question about mpich bandwidth vs message-size plot
by Sajid Ali 27 Nov '19

27 Nov '19

Hi Jeff/Giuseppe, I spoke to the computational specialist from research IT and he confirmed that this particular build of MPICH is indeed broken. As Giuseppe pointed out the hydra launcher was running multi threaded jobs on each node because it was unable to read the nodelist. When we passed a nodelist to mpirun and repeated the same test, the results looked more reasonable (~80us latency instead of ~0.05us due to ch3:TCP). Thanks for the help! -- Sajid Ali | PhD Candidate Applied Physics Northwestern University s-sajid-ali.github.io

1 0

Re: [mpich-discuss] UCX run-time warnings
by Raffenetti, Kenneth J. 27 Nov '19

27 Nov '19

Could you try an development build (mpich@develop) and see if the warnings persist? 3.3.2 only contains critical bug fixes. Our master git branch has diverged quite a bit, especially the ch4 device. Ken On 11/26/19 1:04 PM, Sajid Ali via discuss wrote: > Hi MPICH-developers, > > I’ve built MPICH-3.3.2 from source on my univ. cluster to ensure that > jobs can be launched with slurm’s srun (via —mpi=pmi2, slurm’s native > pmi version). I built mpich from source using spack using ch4:ucx. > > I see the following errors in my outfile (from osu-micro-benchmarks runs): > > |.... slurmstepd: error: mpi/pmi2: no value for key in req .... > [1574612953.443924] [qnode5038:30333:0] mpool.c:38 UCX WARN object > 0x1b88d00 was not returned to mpool ucp_rkeys ... ` | > > as shown in a typical outfile here. > > I don’t see such errors with using openmpi-4.0.2 which was also built > with UCX as the primary transport layer protocol where I compiled a > newer version of UCX from source. > > I’m attaching the build log from mpich install for reference. > > |Windows Terminal[sas4990@quser10 ~]$ spack find -ldv mpich%gcc(a)8.3.0 > ==> 1 installed package -- linux-rhel7-ivybridge / gcc(a)8.3.0 > ---------------------------- btlyk64 mpich(a)3.3.2 device=ch4 +hydra > netmod=ucx +pci pmi=pmi2 +romio+slurm~verbs+wrapperrpath eb74crw > libpciaccess(a)0.13.5 t6h6tgg libxml2(a)2.9.9~python 46i2h6v libiconv(a)1.16 > dnbfmo2 xz(a)5.2.4 xaq5v23 zlib(a)1.2.11+optimize+pic+shared 77kx7yy > slurm@19-05-3-2~gtk~hdf5~hwloc~mariadb~pmix+readline [sas4990@quser10 ~]$ | > > Thanks in advance for your help! > > -- > Sajid Ali | PhD Candidate > Applied Physics > Northwestern University > s-sajid-ali.github.io <http://s-sajid-ali.github.io> > > _______________________________________________ > 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] Question about mpich bandwidth vs message-size plot
by Jeff Hammond 26 Nov '19

26 Nov '19

There is no way for internode latency to be less than ~1 us. Thus, I conclude you measure intranode results with MPICH and internode with everything else. Check your mpirun commands and try again. Jeff On Tue, Nov 26, 2019 at 10:57 AM Sajid Ali via discuss <discuss(a)mpich.org> wrote: > Hi MPICH-developers, > > I’ve been trying to benchmark mpi software currently available on my univ. > cluster with OSU micro benchmarks. I used this script to build and run the > benchmark : > https://github.com/s-sajid-ali/nu-quest-mpi/blob/master/curr_mpi_bench/subm… > > And here is a plot of the results I obtained : > [image: curr_mpi.png] > > Could someone give me a clue as to why the behaviour of mpich bandwidth vs > latency is so odd ? Why does the bandwidth increase first and then fall off > at high message sizes? > > I’m also attaching the config.log for this particular mpich build which > seems to indicate that MPICH was built with ch3 and nemesis:tcp. Does this > mean that it didn’t pick up the infiniband libraries or that it is somehow > using IPoIB? Is there a way to get this information ? > > Thanks in advance for the help! > > > -- > Sajid Ali | PhD Candidate > Applied Physics > Northwestern University > s-sajid-ali.github.io > _______________________________________________ > 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 http://jeffhammond.github.io/

1 0