[petsc-dev] Seeking OLCF users complaining about poor build times

[Barry Smith]

  Thanks. The make time is totally absurd compared to what we get on other systems and the break down between user, sys and nothing is interesting.

  Do you know where /tmp is on the system? Presumably it is fast?

> On Feb 26, 2015, at 8:08 PM, Nathan Collier <nathaniel.collier at gmail.com> wrote:
> No packages / master
> [nate at titan] time python no_packages.py
> ...
> real	14m32.169s
> user	1m51.527s
> sys	3m40.734s
> 
> [nate at titan] time make all
> real	15m56.004s
> user	8m8.971s
> sys	52m42.734s

   Here is what I have so far
                    configure time    make time     Total      compilers     filesystem

Titan                 14m32s         15m56s                                 Intel 14    /lustre/atlas1/geo103/proj-shared/

Mira                   "7m"            "2m"                     IBM        /gpfs/mira-home/

Hopper

Linux laptop                                                    Gnu 

Apple laptop         1m14s             54s         2m8s         clang         compile and compiler local

Linux workstation    1m11s             22s         1m33s        Gnu        compile and compiler local
                     1m37s             29s         2m6s         Gnu         compile directory local; compiler directory remote
                     3m11s             25s         3m36s       Intel 13     compile directory local; compiler directory remote

> 
> [nate at titan] cat no_packages.py
> 
> #!/usr/bin/python
> configure_options = [
>   '--known-level1-dcache-size=16384',
>   '--known-level1-dcache-linesize=64',
>   '--known-level1-dcache-assoc=4',
>   '--known-memcmp-ok=1',
>   '--known-sizeof-char=1',
>   '--known-sizeof-void-p=8',
>   '--known-sizeof-short=2',
>   '--known-sizeof-int=4',
>   '--known-sizeof-long=8',
>   '--known-sizeof-long-long=8',
>   '--known-sizeof-float=4',
>   '--known-sizeof-double=8',
>   '--known-sizeof-size_t=8',
>   '--known-bits-per-byte=8',
>   '--known-sizeof-MPI_Comm=4',
>   '--known-sizeof-MPI_Fint=4',
>   '--known-mpi-long-double=1',
>   '--known-mpi-int64_t=1',
>   '--known-mpi-c-double-complex=1',
>   '--known-sdot-returns-double=0',
>   '--known-snrm2-returns-double=0',
>   '--known-mpi-shared-libraries=0',
>   '--with-ar=ar',
>   '--with-cc=cc',
>   '--with-clib-autodetect=0',
>   '--with-cxx=CC',
>   '--with-cxxlib-autodetect=0',
>   '--with-fc=ftn',
>   '--with-fortran-datatypes=0',
>   '--with-fortran-interfaces=0',
>   '--with-fortranlib-autodetect=0',
>   '--with-ranlib=ranlib',
>   '--with-shared-ld=ar',
>   '--with-clean=1',
>   '--with-c2html=0',
>   '--with-shared-libraryies=0',
>   '--with-debugging=0',
>   '--with-x=0',
>   '--with-batch=1',
>   '--with-mpi-dir=/opt/cray/mpt/6.3.0/gni/mpich2-intel/130',
>   'COPTFLAGS=-O3 -msse3 -fPIC ',
>   'CXXOPTFLAGS=-O3 -msse3 -fpic ',
>   'FOPTFLAGS=-O3 -msse3 -fPIC ',
>   'FCOPTFLAGS=-O3 -msse3 -fpic ']
> 
> I can add packages too, just let me know if that is still interesting.
> 
> Nate
> 
> 
> On Thu, Feb 26, 2015 at 7:35 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> 
> > On Feb 26, 2015, at 6:26 PM, Jed Brown <jed at jedbrown.org> wrote:
> >
> > Barry Smith <bsmith at mcs.anl.gov> writes:
> >>  Sure it will be 2 minutes on the laptop and 7 minutes on the big
> >>  system but as I said in my other email that is acceptable and no one
> >>  will listen too us (for good reason) if we bitch about it. It is
> >>  when it is 4 minutes on the laptop for configure and build and 1
> >>  hour on the big machine for configure and build that we have the
> >>  right to scream and yell.
> >
> > Yeah, that Mira time is pretty good, though the login nodes are a
> > different architecture from the compute nodes (POWER vs BG/Q) so one
> > would typically use batch, in which case configuring goes up to 20-30
> > minutes of human time.
> 
>   We could have a script that runs configure; submits the result to queue; gets result from queue; runs the new configure then runs make. Assuming an instant queue turn around and that both configures take the same amount of time it means the configure is 14m; not great but still hard to bitch about too much.
> 
> 
> 
> 
> <no_packages.log>