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

[Satish Balay]

On Thu, 26 Feb 2015, Victor Eijkhout wrote:

> It's a 16-core beast, not much different from the compute nodes:

> >  BTW: PETSc does a parallel compile (which is why it is so fast) did the sysadmin complain ?
> 
> No he didn't but I don't want to tempt him. I'll talk to him tomorrow.

PETSc configure defaults to using a MAKE_NP smaller than the available
node_core_count (for node_core_count > 4) - this way - it avoids
swamping the whole machine [and try to avoid sysadmin warth and
such..]

In your case - with 16 cores available - PETSc configure scheduled
'make -j 13' job - and that ran for a bit more than a minute. And
'user'+'sys'/'real' times give a factor of 6.2 - so thats equivalent
to keeping 6 cores 100% busy?

BTW: We'd like to do the same for all externalpackage builds [during
configure stage]. And if sys-admins object - you can reduce the load
by tweaking the configure option --with-make-np :)

Satish


<from previous e-mail>
=========================================
gmake[1]: Leaving directory `/work/00434/eijkhout/petsctest/petsc-3.5.3'
Now to check if the libraries are working do:
make PETSC_DIR=/work/00434/eijkhout/petsctest/petsc-3.5.3 PETSC_ARCH=arch-test test
=========================================

real    1m13.661s
user    5m47.502s
sys     1m44.092s