[Beowulf] HPC workflows

[John Hanks]

On Fri, Dec 7, 2018 at 7:20 AM John Hearns via Beowulf <beowulf at beowulf.org>
wrote:

> Good points regarding packages shipped with distributions.
> One of my pet peeves (only one? Editor) is being on mailiing lists for HPC
> software such as OpenMPI and Slurm and seeing many requests along the lines
> of
> "I installed PackageX on my cluster" and then finding fromt he replies
> that the versiion is a very out of date one delivered by the distribution's
> repositories.
>
> The other day I Was interacting with someone who was using a CentOS 6.5
> cluster on the Julia discussion list. His cluster uses the original SGE
> version.
>

This is one of my long term pet peeves, I call it the "IT Drone Stable
Software Release Delusion". It manifests itself as "version compatibility
matrices" and a side effect are things like RHEL 5 clusters that "can't be
updated" but have a massive software stack where pretty much everything in
the OS has been manually rebuilt by hand, up to and including the kernel,
to get the latest versions. These clusters become ideal places to run
containers because that's the only way to get a modern OS past the sysadmin
and onto the cluster.

The delusion is usually strongest in places where there are people who
justify their existence via "Change Management/Change Control" meetings,
but it can creep into any environment in subtle ways like "we never run .0
releases..." This delusion dovetails into an underlying fear of change,
sees the word "freeze" bandied about a lot and ultimately leads to
environments where significant amounts of pain are passed on to the users
while the IT drones (or IT drone apprentices) study the ancient scrolls of
compatible version matrices and try to get or maintain ITIL certification
(with bonus points achieved if they are wearing a six sigma black belt).

griznog
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.beowulf.org/pipermail/beowulf/attachments/20181207/baeaf180/attachment.html>