iiinova — where your digital self keeps evolving

Glenn K. Lockwood @glennklockwood.com · 23/09/2026

Spent a few months preparing the tech docs for DataEnclave, VAST's new confidential computing runtime. W/ help from VAST R&D, I learned how attestation etc worked well enough to describe and defend it. Short version: www.vastdata.com/blog/introdu... 15-pager: www.vastdata.com/resources/wh... #AI

vastdata.com

Introducing VAST DataEnclave Confidential AI

VAST DataEnclave runs AI workloads in hardware-isolated environments, so proprietary models and sensitive data can meet on untrusted infrastructure.

031

Glenn K. Lockwood @glennklockwood.com · 21/09/2026

I wrote this doc to explain KV caches and KV cache offloading many months ago. I found salespeople couldn't confidently tell when their customers might have a problem, because all the content online is straight marketing or way too in the weeds. blog.glennklockwood.com/2026/09/what...

blog.glennklockwood.com

What are KV caches really?

KV caching lies sufficiently deep within the guts of how transformers work that really understanding how it can be used goes beyond any si...

161

Glenn K. Lockwood @glennklockwood.com · 17/09/2026

Malaysia is allegedly evaluating Huawei's Ascend 910C for their sovereign AI buildout. If Malaysia picks Huawei over NVIDIA, it'd be the first big buy of these chips outside of China. Egypt also name-dropped as a nation being pursued by Huawei. www.freemalaysiatoday.com/category/hig...

freemalaysiatoday.com

Malaysia weighs Huawei chips for sovereign AI push

The country is seriously considering using Huawei’s AI hardware as the backbone of a RM2 billion initiative aimed at giving Malaysia greater control over national data.

000

Glenn K. Lockwood @glennklockwood.com · 10/09/2026

The IO500 committee just stripped Sugon of its #1 position in the Production list because it appears that their ParaStor F9000 isn't a real product, or it could not be verified as one. Verify here: io500.org/list/isc26/p... My notes on ParaStor: glennklockwood.com/garden/paras... #HPC #notreally

io500.org

IO500 - ISC26 - Production List

062

Glenn K. Lockwood @glennklockwood.com · 03/09/2026

I recently had the opportunity talk to @nicolehemsoth.bsky.social about a few failures in #HPC and #AI infrastructure that taught me some important lessons about designing systems and managing risks. Have a look, have a listen. Hope others find it interesting! www.vastdata.com/blog/failure...

vastdata.com

Failure is a Finding: The Hard Lessons Behind Computing's Biggest Bets

VAST delivers the first AI Operating System, unifying storage, database, and compute to drive agentic computing and data intensive workloads. Learn more.

063

Glenn K. Lockwood @glennklockwood.com · 01/09/2026

MLPerf Storage results are out and StorageReview pointed out the interesting disconnect between MLPerf leaderboard winners and companies with the largest #storage footprints in the #AI clouds. Suggests this benchmark may not reflect real needs. www.storagereview.com/news/mlperf-...

storagereview.com

MLPerf Storage v3.0: 877 GiB/s Checkpoints, a Cloud First, and a Leaderboard Turned Over

MLPerf Storage v3.0 results: 877 GiB/s checkpoints from Everpure FlashBlade//EXA, new KV cache and vector DB tests, 143 submissions.

140

Glenn K. Lockwood @glennklockwood.com · 23/08/2026

Just reflects my lack of familiarity with Flux. Most of what I encounter in the wild is Slurm and Airflow, both of which suffer the same scalability limitations.

000

Glenn K. Lockwood @glennklockwood.com · 22/08/2026

I recently wrote up an argument about why batch schedulers like Slurm are architecturally insufficient for autonomous scientific laboratories and agentic workflows. Event-driven orchestration is the scalable alternative. www.vastdata.com/blog/autonom... #HPC

vastdata.com

Autonomous Science is Inheriting a Workflow Infrastructure Problem

VAST delivers the first AI Operating System, unifying storage, database, and compute to drive agentic computing and data intensive workloads. Learn more.

170

Glenn K. Lockwood @glennklockwood.com · 13/08/2026

Funny, there’s a reason NERSC took its datacenter out of this location. It’s not a great place to put one. www.datacenterdynamics.com/en/news/form...

datacenterdynamics.com

Former LBNL supercomputing facility in Oakland, California, targeted for 20MW data center

Behring looks to turn vacant site back into live data center

132

Glenn K. Lockwood @glennklockwood.com · 13/08/2026

I have a theory that every panel about “how AI will affect XYZ” will devolve into nonspecific discussion around “here is where the AI tools I use today fail” and “how do we trust AI?” regardless of what XYZ is.

020

Glenn K. Lockwood @glennklockwood.com · 13/08/2026

This is the diagram I’m most proud of. Stacking ~700 different LLM training jobs observed from the VAST fleet on a common timescale to figure out whether there is a “feeding the GPU” problem anywhere close to the write intensity of periodic checkpoints. There is not. Only read burst is at job start.

022

Glenn K. Lockwood @glennklockwood.com · 12/08/2026

The lack of shame some people have in copypasting prose straight out of Claude into Powerpoint is shocking. Were these people just recycling their students’ slides before, and now just tap a different source of material?

141

Glenn K. Lockwood @glennklockwood.com · 09/08/2026

Problem is, it's not my day job to do this kind of analysis. So here I am, frantically banging on signal analysis and statistical models on a Sunday, to have something compelling to say 😩 As with all longitudinal workload analysis, people are messy and hard to understand or predict.

120

Glenn K. Lockwood @glennklockwood.com · 09/08/2026

I am attending the ModSim conference for the first time next week, and I'm presenting analysis and design modeling of I/O performance requirements for frontier model training based on a year's worth of production data across the VAST hyperscale fleet (>900 training jobs, >140K GPU-years of compute)

221

Glenn K. Lockwood @glennklockwood.com · 07/08/2026

I recently upgraded to Claude Max and started using Opus 5 for exploratory analysis. If this is what the AI community things approaches "PhD-level capability," I don't know what to say. It's sloppy, inconsistent, and walks back conclusions at the lightest of questioning.

110

Glenn K. Lockwood @glennklockwood.com · 04/08/2026

Surprised to see a solution so heavily focused on tiering to HDD for capacity now tier to…more HDD for capacity. Curious what Wasabi does that PanFS doesn’t (and vice versa) to make this pairing complementary.

010

Glenn K. Lockwood @glennklockwood.com · 01/08/2026

yes you’re right; it reduces compute. it does also open another way to parallelize so you can effectively aggregate more GPUs’ HBM, but it doesn’t reduce the need for HBM capacity as I erroneously wrote.

000

Reposted by Glenn K. Lockwood

v @vsoch.bsky.social · 30/07/2026

I gave the invited talk for the Systems track at #PEARC26. If you were not able to attend, you're in luck! 🍀 I walked back to my room and recorded a second time so all would have access in the present and into the future. youtu.be/qiKfTgYoKkE?...

youtu.be

🌀 The Agentic HPC Center: Orchestrating the Future of Science

YouTube video by vsoch

152

Glenn K. Lockwood @glennklockwood.com · 29/07/2026

I’m glad someone liked it! Hard needle to thread in crafting a technical plenary that is interesting, accessible, and thematic to “resilient roots.” Also relieved that my SDSC colleagues didn’t mind me spilling the tea on one of Gordon’s failures. #HPC #alittleAI

151

Glenn K. Lockwood @glennklockwood.com · 29/07/2026

Nobody came 😩 Just kidding. Was mic check. Here’s the actual crowd.

030

Glenn K. Lockwood @glennklockwood.com · 28/07/2026

It’s my first PEARC. I was at XSEDE’13 but one of the PEARC chairs told me that doesn’t count. No swag from me; I’m actually attending to give the Wednesday plenary (pearc.acm.org/pearc26/plen...) which will have nothing to do with VAST.

pearc.acm.org

Plenary Speakers - PEARC26 - Resilient Roots + Empowered Communities

Meet our Plenary Speakers Tuesday, July 28 – 9:00 am – 10:15 am Opening Ceremony Keynote: Amanda RandlesDuke University Amanda Randles is the Director of the Center for Computational and Digital Healt...

000

Reposted by Glenn K. Lockwood

parasharmanish.bsky.social @parasharmanish.bsky.social · 22/07/2026

Another great addition to Utah’s amazing pro-human AI innovation ecosystem built on public-private partnerships.

011

Glenn K. Lockwood @glennklockwood.com · 21/07/2026

Surprising to see a software company go into the hardware business, selling an unusual 2U70 dual-controller flash server. It's hard to source flash these days, so there are strong market incentives to ship software on whatever hardware your customers can find.

000

Glenn K. Lockwood @glennklockwood.com · 15/07/2026

We are now in the age of AI-generated conferences. Today’s keynote by Gropp prominently features distinctly Claude-generated plots; yesterday had an entire presentation delivered by an AI voiceover of slides. And slides are full of Claude design tells.

060

Glenn K. Lockwood @glennklockwood.com · 15/07/2026

GPUs only, no storage. So no problem.

100

Glenn K. Lockwood @glennklockwood.com · 14/07/2026

First time I’ve seen someone with a PhD actually predict that data centers in space will be a real thing. Granted, it’s a CS PhD, not a physical sciences one. #HPC #notreallythough

2223

Glenn K. Lockwood @glennklockwood.com · 10/07/2026

Infinia hardware should be cheaper since it doesn’t require HA pairs. Software is software and can be however much or little as your sales rep likes you.

010

Glenn K. Lockwood @glennklockwood.com · 09/07/2026

With Infinia gaining support for files, is there a reason for anyone to buy EXAScaler anymore? Why would someone choose Lustre over Infinia if Infinia seems to support everything that Lustre does but with enterprisey glitter? Curious to see how DDN positions these two lines. #HPC

210

Glenn K. Lockwood @glennklockwood.com · 09/07/2026

Block off your calendar for Wednesday at 10am! I'll be there.

041

Reposted by Glenn K. Lockwood

OGAWA, Tadashi @ogawa-tadashi.bsky.social · 09/07/2026

=> Architecting for Frontier AI Labs: How Anthropic scales Claude on TPUs, Google & Anthropic, Google Cloud Next 2026, Apr 22 youtube.com/watch?v=LEPY... content-cdn.sessionboard.com/content/3e8V... Automated fault recovery & intelli scheduling to maintain continuous uptime across massive pod-scale

111

Glenn K. Lockwood @glennklockwood.com · 08/07/2026

Yes I am giving the technical plenary at PEARC. More details should be publicized soon!

010

Glenn K. Lockwood @glennklockwood.com · 08/07/2026

I wrote up some notes after my week at #ISC26: blog.glennklockwood.com/2026/07/isc2... Interested to hear others' perspectives on highlights and things I may have missed. #HPC

blog.glennklockwood.com

ISC'26 recap

Last month was the 2026 ISC High Performance Conference in the beautiful (and sweltering) Hamburg, Germany. It was my fifth time att...

3102

Glenn K. Lockwood @glennklockwood.com · 02/07/2026

I somehow missed that China has developed its own implementation of InfiniBand, called ScaleFabric 400, to compete/supplant NVIDIA/Mellanox IB. Oddly enough, its manufacturer (Sugon) is not a member of IBTA though. en.eeworld.com.cn/mp/EEWorld/a... #HPC

en.eeworld.com.cn

The first domestically produced InfiniBand has been launched; real-world test data tells you just how powerful its performance is.-Electronics Headlines-EEWORLD

Recently, another technological high ground long monopolized by foreign countrie......view more！

031

Glenn K. Lockwood @glennklockwood.com · 26/06/2026

Interesting that C-DAC, whose indigenous Indian interconnect Trinetra is based around libfabric, opted to implement verbs support to make Lustre work instead of adopting Cray’s kfabric LND. #ISC26

010

Glenn K. Lockwood @glennklockwood.com · 26/06/2026

1. General LLMs beat coding-specific models at writing Fortran, likely bc code models are overfitted to Python 2. This conf begins the age of Claude-generated pptx. Clear tells 3. My phone doesn’t like these #ISC26 projectors

160

Reposted by Glenn K. Lockwood

Robert Henschel @roberthenschel.bsky.social · 26/06/2026

Now happening at #ISC26 #HPC

011

Glenn K. Lockwood @glennklockwood.com · 25/06/2026

He has a chance to mention the Ozaki scheme but instead cites a 20-year-old algorithm that is the problematic core of HPL-MxP.

120

Glenn K. Lockwood @glennklockwood.com · 25/06/2026

Don’t trust any of them. If the source is a tweet from a guy who has a history of having a loose relationship with the truth, then it shouldn’t be in a conference keynote.

020

Glenn K. Lockwood @glennklockwood.com · 25/06/2026

This is the worst Jensen keynote ever

020

Glenn K. Lockwood @glennklockwood.com · 25/06/2026

I wonder if ground has been broken on this datacenter yet. Because it takes two years to build such a facility without the federal land red tape part. How commercially meaningful will Blackwells be in two years?

130

Glenn K. Lockwood @glennklockwood.com · 25/06/2026

Funny that “software adapt to whatever hardware shows up” is exactly how everyone’s favorite AI models are developed. Model architecture iterates much faster than chip design. This only works if your problems stand still.

200

Glenn K. Lockwood @glennklockwood.com · 25/06/2026

How many wrong statements can you fit into a single sentence? Holy cow.

220

Glenn K. Lockwood @glennklockwood.com · 25/06/2026

I can’t wait to hear what Jack Dongarra thinks about the future of #HPC at #ISC26’s closing keynote.

120

Glenn K. Lockwood @glennklockwood.com · 25/06/2026

Dumb stuff that results in weird requirements in #HPC RFPs. Ultra dense GPU racks with all liquid requires unusual datacenter construction considerations. #ISC26

362

Glenn K. Lockwood @glennklockwood.com · 25/06/2026

Ogawa-san is how I first came to know about China’s new exascale system a while ago, and I assembled a quick spec page a few weeks back (glennklockwood.com/garden/syste...). Notably has a BIG liquid cooled storage system (67 racks, 10 TB/s).

glennklockwood.com

LineShine (NSCC-SZ)

LineShine is an all-CPU exascale supercomputer at the National Supercomputing Center in Shenzhen (NSCC-SZ), built entirely from domestically produced Chinese hardware with no reliance on foreign chips...

051