kincl

The country code GB in some of the tables should show the source economy being Great Britain right? Am I misunderstanding the table?

If one of the main advantages is being able to suspend to disk why not just implement CRIU (Checkpoint/Restore In Userspace) in Kubernetes?

Great thread! I noticed a bunch of the comments are from Sandia/LLNL/LANL all of which are mostly focused on the National Nuclear Security Administration side of the Department of Energy which is focused on the various aspects of maintaining the nuclear stockpile of the US.

I worked at Oak Ridge National Laboratory in High Performance Computing and did not work with anything directly nuclear at all. The HPC efforts of the DOE are under the Office of Science (separate and at the same level as the NNSA) which is focused on more wider scientific impact and application than just nuclear. The Office of Science has a number of program offices that focus on all different kinds of science from basic energy sciences/physics to biological/environmental and scientific computing (where HPC is funded in DOE).

I agree that the work/life balance is great and it is definitely a slower pace than what you would find in industry. The lab system is huge and there are plenty of opportunities but on the Office of Science side I like to break it down between what I think of as a research group and a user facility.

Working in a research group is much like academia, they mostly require a PhD and from what I could tell performance is judged on publication output. These folks also write grant proposals that come from DOE program offices for funding their own research. In some cases I have seen these groups employ non-PhDs to be computational scientists and write code.

The user facilities are long-running projects funded by the DOE at the labs to provide specific capabilities to researchers, sometimes just for DOE scientists but a number of them are open to scientific researchers all over the world. This is where I have the most experience where I worked at ORNL's National Center for Computational Sciences on the Oak Ridge Leadership Computing Facility (OLCF). These projects are generally well funded and have all kinds of interesting challenges to solve. For example, the OLCF has consistently deployed the number one supercomputers on the Top500 list and it offers those computational resources to anyone through their allocation program INCITE which supports many different computational modeling and simulation experiments. Other examples of user facilities at ORNL are the Spallation Neutron Source and the High Flux Isotope Reactor.

One thing I have noticed since moving from ORNL to industry is that the sense of shared purpose does not extend as far in the lab system as it does in company. What I mean is that with the small research group and with a user facility like the OLCF there is shared purpose with the people in those groups but it does not go much beyond that. A lab is generally made up of lots of different research groups and a few facilities but beyond the drive for "Science!" there is not a lot of shared purpose or collaboration at a macro level. The analogy I use is that a lab is a bunch of small dinghy boats that are all generally moving in a similar direction but a company is a single ship with a specific purpose driving it forward.

Overall I loved my experience at ORNL, I learned so much working with so many smart people and made friends that I will have for life.

I just realized that OpenBSD unbound was likely named as a counter to ISC BIND. Similar to Sun Java and the Eclipse Foundation.

I think KubeVirt fits better in an environment that is already running Kubernetes. We have found it useful for running some of our heavyweight CI jobs that have virtualization requirements because we can manage the VMs in the same way that we manage our containers.

In general though we have only used KubeVirt for ephemeral workloads.

Yep, actually all of our user projects are unclassified at the OLCF.

Yep, we have battery-backed generators for UPS and a transfer switch at the 480-V feed that comes into the room but it is not enough to power the compute nodes. The UPS allows cluster management nodes and the parallel filesystem (which is a small cluster by itself) to ride through full outages and other PQE.