Recent searches
Search options
For at least a decade, US higher ed has been all-in on outsourcing as much IT to Google and MS as possible. It's been years since the last major US university replaced internally-operated email with one of those two.
One of the few computer-related services not yet completely surrendered to "the cloud" is HPC, but that's not for lack of trying. Years of literally screaming lies into the faces of researchers and IT people has taken its toll - a generation quit/retired/laid off.
To me, this all amounts to en-masse institutional self lobotomization. The talent that created the collection of technologies we call the Internet were replaced by service contracts with monopolists - contracts negotiated by incompetent poseurs. The result is the complete loss of self-determination - you get what they let you have.
The old team isn't getting back together, but efforts like Othernets might help us field the team we need now.
@lolcat Could not agree more. Been in DOE HPC for over 20 years, and yikes the things that I am seeing.
At the risk of coming off all conspiratorial, from where I sit the Big Data people and the Cloud people were a near union set, and their main motivation was turning the data collected about us by the hyperscalers into $$.
Back in 2012, they didn't know the result would be gen AI. In fact, one of their loudest claims back then was: "It's not about the cycles, it's about the data!!!!!" Which was as nonsensical as it sounds.
Of course, now we have MS buying nukes to run those GPUs.
As I sit here finishing a tuna salad sandwich and a beer, I recall at least 3 occasions between 2012 and 2017 in which our regional DoE lab, PNNL, with the support of university admin and one especially powerful and troublesome CS chair, simply asserted that they were taking over our on-prem HPC operation. In one case, they even issued a press release or two, which led to stories in PNW press.
They only failed because (at the risk of flattering myself) I intervened, and...
... because our HPC operation was owned and governed by the users.
It turns out that it was a critical miscalculation on the part of central admin and IT to refuse to support centralized HPC. That forced the research community to cooperate to create a system largely outside central control. So, instead of DoE just needing sign off from the Vice Provost for Research, they were confronted by hundreds of researchers with tens of million$$ of their own money invested.
...
...
It also helped that HPC was the only customer propping up the new, on-prem data center.
Years earlier, the Cloud Mafia, before they identified as such, had successfully lobbied the state legislature for million$$ to build the facility. By the time engineering work began, they knew they'd f'd up, and did everything they could to cripple the facility for HPC. Nonetheless, they knew they'd look stupid if the facility remained empty, so we had that in our back pocket.
As long as I'm way out here alone on this limb, I'll also say that DoE positioned people as faculty at large regional public universities specifically for the purpose of preventing any on-prem HPC. As soon as it was clear they'd failed, these people disappeared back into the national labs.
I also regard one former PNNL HPC leader, long since sequestered at a different national lab, to be a national security risk (angry drunk with a security clearance)
It is a weird, weird world.