On Mon, Aug 29, 2016 at 09:11:50AM -0500, Jason Bacon wrote:
2) Roughly how long does it take to do a bulk build of the entire tree on
a single high-end server? I was thinking of doing builds on one of our
production clusters, but I'm leaning against it due to some issues with
leakage of Yum packages into certain builds. I think it would be safer to
do the builds on a dedicated server with a minimal CentOS installation.
Define high-end server :) I have a ~2 year old dual Xeon (8 "real" cores
in total) with 64GB RAM that I use for the clang bulk builds of current.
That can do a full build in about 2 days, give or take. With enough RAM,
you can do a build entirely within tmpfs for the non-persistent data.
I wouldn't worry too much about the base installation in the cluster,
but follow an approach similar to what I am using for all my builds.
Essentially, prepare a reasonable minimalistic CentOS installation,
including gcc and gfortran. Create a tarball from what and use it to
chroots. You'll likely need some additional mount points for /sys, /proc
and /dev, but you will end up with an image that is nicely decoupled
from the actual base installation.
Feel free to contact me (off-list) for further details.
Joerg