v

v

@vsoch.bsky.social

Principal Computer Scientist https://vsoch.github.io πŸ«’

555 followers29 following355 posts

Upper Trouble Maker, Class III+ 😎 An amazing trip! Not enough rapids for me πŸ˜‰

This is the first laboratory photograph I have ever been in, and that feels pretty great. Can you find me? lnkd.in/p/gt5GrSh5 Thank you to Sandia and the other labs for putting together this event! πŸ₯³

@quantummachines.bsky.social do sites provision your interface, the FrontendServicer? We are prototyping with the #FluxFramework graph scheduler, and it occurred to me that this work might be useful for others that control that interface. Are there folks I can talk to, community meetings? Thank you!

πŸ«’Olive oil update! I’ve finished 69 bottles in about 5 months, and my current favorites are La Tourangelle Smooth and Fruity (on my second tin), Finca La Barca (finishing my second bottle) and Castillo de Canena Sherry Cask (fourth bottle). πŸ˜‹

I present the two-queue problem for co-scheduling of two different queues, and design solutions using the #FluxFramework scheduler Fluxion in Kubernetes and (WIP) on-premises. Our graph-based approach models quantum as a first class resource, and is extendable to other custom types

I get obsessively interested in topics, to the point my curiosity is insatiable. It is happening with quantum computing, and here is some learning.😊 vsoch.github.io/2026/quantum... I think along with scheduling work I want to model scheduling algorithms with qiskit (and similar).

v reposted

🌱 #ACM #PEARC26 Invited Speaker | Systems Track Join Vanessa Sochat (vsoch.bsky.social), Principal #Computer Scientist at Lawrence Livermore National Laboratory, as she explores the future of #agentic #HPC! 🎀 The Agentic HPC Center: Orchestrating the Future of Science πŸ—“οΈ July 28, 2026 | 11:00 AM CT

Want to learn about #FluxFramework with live presentations and tutorials? Come to our HPCIC Developer Tutorial, July 21st! I may have something up my sleeve this year that is a bit cutting edge... hope to see you there! llnlfed.webex.com/weblink/regi... Registration is free and open to anyone!

πŸ«’ Olive oil review! I loved this oil because it tastes like arugula. I also do a deeper dive on the polyphenol Oleocanthal, and how it is works as an anti-inflammatory and interacts with TRPA1 and COX enzymes. Enjoy! πŸ’ƒ olive-oil-girl.blogspot.com/2026/05/oliv...

olive-oil-girl.blogspot.comπŸ‡«πŸ‡· Oliviers & Co. from Mane, FranceI finished one of my favorite olive oils this last week, and with a little help from a tasting that I hosted with a small group. This is La ...

Interested to learn about our work in agentic science, with some added #FluxFramework? Please attend the #HPCKP meeting in Barcelona! πŸ’ƒ I am excited for this work, and for the future of science. After recording this talk, I was moved to dance!🌹 hpckp.org/annual-meeti...

I am building large containers locally that have layers too large for push from my home wireless. "Ah, I'll use an instance on Google Cloud!" I thought. I tried to get *one* c4-standard-96 instance, and then c3-standard-88. Could I? Nope. I tried every zone. 😭 It is Saturday. πŸ™ƒ

An on-demand model does not work when you are not the target customer, and the demand is greater than the supply. I am using an on-premises node. The same interaction is harder. I needed to find and COPY three custom certs to my container, and use rootless Podman. I wish Google Cloud worked for me.

An agent just wrote and ran an entire #snakemake workflow using our agentic tools + mcpserver! 🐍 Running snakemake would be too easy. I've given the agent just inputs + a wrappers catalog and said "figure it out." The agent gets controlled access to a workdir, and incrementally figures out steps.

GarcΓ­a de la Cruz from Toledo Spain. What was most interesting was learning about the history of the amphora. They are used to date shipwrecks. At first glance the design seems nonsensical, but this is with my applied bias of storage today. Enjoy! olive-oil-girl.blogspot.com/2026/04/garc...

olive-oil-girl.blogspot.comπŸ‡ͺπŸ‡Έ GarcΓ­a de la Cruz from Toledo SpainGarcΓ­a de la Cruz is a premium olive oil produced in Toledo, Spain, and founded in 1872. They advertise as one of the oldest oil mills in Sp...

New olive oil review! πŸ«’ These are made by the same producer in Morocco. Both have good price points, and beautiful bottles. I recommend Desert Miracle if you want to try oil that is subtly peppery, and Atlas if you want something primarily smooth. olive-oil-girl.blogspot.com/2026/04/atla... Enjoy!

olive-oil-girl.blogspot.comAtlas and Desert Miracle Moroccan Olive OilsI present to you two elegant, dark bottles. One is octagonal with shiny, embossed gold lettering, and an elegant "A" with what appears to be...

Running jobs with our #Slurm Operator today! 😘 I'm enjoying learning about build, admin, and config of Slurm. I'm running on AWS hpc7g with EFA, and (so far) the jobs are comparable in performance to Flux! Would ❀️ to write a paper, "Workload managers, they are all pretty good. Stop the quibbles" πŸ˜†

View full thread

My design approach is simplicity. You install the operator with two commands (eventually will be one) and create your cluster. The spec I might change as I learn more about Slurm. Here is the cluster for EFA. This could be reduced to a node.nodespec and size for a simple deployment.

I am also wanting to better expose nuanced sections of the slurm.conf so everything is easy to change. I have a lot of learning (and work that I would like to do, since my style has changed since 2023) and I mostly want to say I am enjoying working on this. That's all folks! Back to experiments. πŸ€“

Running jobs with our #Slurm Operator today! 😘 I'm enjoying learning about build, admin, and config of Slurm. I'm running on AWS hpc7g with EFA, and (so far) the jobs are comparable in performance to Flux! Would ❀️ to write a paper, "Workload managers, they are all pretty good. Stop the quibbles" πŸ˜†

Why was it so hard to make a cluster? I tried AWS Parallel Cluster and PCS, and I think they would work if I wanted a "vanilla out of the box" cluster, but I want to customize it. I tried for about 16 hours. It was arduous to have layers of configs, whether in CloudFormation, helm, or for pcluster.

I got to the point where I needed something that would deploy easily and quickly. I updated our operator (created 3 years ago, before SUNK and Slinky) for ARM support and it's working great! 😘 I am going to be removing the dependency on JobSet.