I think you have some fundamental misunderstandings here. First off, there absolutely is an amount of "hot spares and RAID" that would save you from "failure rates that high", and ultimately hot spares aren't going to help anyway -- either you're actively running things in which case you need enough hot spares to handle disk replacements until the next time someone opens up the computer in question to add more disks, or at the opposite extreme you need enough active redundancy to survive a random 1/3 of disks failing. With that said, it sounds a lot like you think the model here is "a bunch of drives sitting on a shelf", which isn't how anyone sane would run this type of service. The Backblaze model (at least as of 2016 when they last published POD designs) has all disks actively connected to computers. Those 2016 designs heavily use expanders which means that you're not going to get full bandwidth to all disks at once, but it's not like the disks are powered down and inaccessible. It almost certainly makes sense to use spare disk I/O bandwidth to run some frequency of scrubbing of existing data -- if only else to figure out quickly when disks fail so you don't need to keep so much redundancy around. Like, think of this as weeks to months between scrubs that verify that all backed up data is healthy, not accessing cold data after disks sat on a shelf for years.

A fundamental problem with "remote controlled by the cheapest of workers somewhere on earth" is latency. Just considering network latency, you're looking at hundreds of milliseconds of round trip time if you try to have (for example) people in India remote controlling a vehicle in America.

That might be fine for high level executive directions for an AI system, but it isn't going to work if you want them to actually take over and drive the vehicle. The latter really needs someone probably within a few hundred miles.

The main point of a support structure like the one apparently negotiated here is exactly to provide enough time for a migration to happen, should VMWare or its successor company/companies decide to dramatically change the terms/prices of the products in question.

Like, if VMWare suddenly got 100x more expensive, that's fine -- they'll just move away to something else, but such a migration takes time so they want locked in non-price-gouge rates for enough time to actually do that migration. Additionally, having such a support guarantee would make it more possible to take advantage of VMWare specific features that would make it more difficult to migrate to something else.

The problem is that Broadcom and/or the supplier appear to have reneged on at least the spirit of that deal.

I assume the "just switch" will very likely happen. The point is to give Tesco enough time to actually do that switch.

I am just slightly confused here. You're talking about "defending violent criminals" while the Trump administration itself appears to be mysteriously reneging on their own promises to release details about a notorious child sex trafficker with close historical ties to Trump, even while that same administration mysteriously moves his closest accomplice to a minimum security prison.

No amount of deployment of national guard to DC is going to solve the criminal gang problem if the people issuing their orders aren't even pretending to not be regulars at that Pizzagate place.

Castle Bravo is a somewhat odd example to bring up in the context of Lithium 6 versus 7: Lithium 7 fissioning from high energy neutrons and producing tritium was the reason that Castle Bravo was 2.5x larger than expected.

I'm not sure what fantasyland Japan you're living in, but 10 gigabit fiber service is nowhere near as ubiquitous as you're saying. It certainly exists, but it isn't "the basic fibre service" -- it is a still somewhat limited area availability product, and unless something has changed in the past year or so it is a premium product that isn't the lowest tier of any provider.

Why is the first mention of the specific "galaxy next door" (the Large Magellanic Cloud) at the bottom of the summary?

This is supposed to be news for nerds. We know what the Large Magellanic Cloud is. Meanwhile the "summary" buries this information further in its text than TFA!

This isn't the issue. Neo4j didn't violate FSF's copyright.

You distribute your top level license that includes the original text of AGPL alongside some additional restrictions. This is a 100% expected thing to do. The original license has in section 7 a permitted set of restrictions, and then a clause saying that if you find any other "further restrictions" you can strip them out.

The problem is that Neo4j actually did that -- they added an "other further restriction", and someone went and distributed a version with the restriction stripped out as the license says they are allowed to do.

Neo4j could have distributed a version of AGPL with section 7 modified -- but then they actually would be violating the copyright on AGPL. Alternatively they could have written their own license, but then they'd have to pay lawyers to write a proper license that properly expresses what they want which is tricky/expensive to get right.

Do these random contracts have specific language in them saying you aren't allowed to distribute modified copies of their text?

It's one thing to fair use handwave-as-conventionally-accepted only-copyright-violation-in-a-hypertechnical-sense copying text of random licenses and contracts, but a typical conventional contract or license doesn't specifically ask you not to distribute modified copies of it.

It isn't that creative. You've been given a license that has a certain set of terms, one of which says that you're allowed to ignore additional attached terms.

This is ultimately a copyright issue -- of the license itself. FSF doesn't want their licenses used for non Free Software contexts, which is why you are only allowed to distribute the license unmodified and said license text is poisoned against trying to attach additional stuff to it.

From the point of view of FSF this wouldn't primarily be about anything involving Neo4j licensing -- at the end of the day, assuming they own copyright on any relevant code and aren't relying on (A)GPL to have a valid license to distribute a derivative work of other (A)GPL work, they're allowed to write their own license with whatever text they want. The question here is whether people are allowed to rip off the FSF's license texts to license their nonfree software.

While that is largely true, there are definitely scenarios that get more complicated notably if the account holder is actually owed money. There's an extremely relevant issue (mentioned in TFA) with Revlon loans back in 2020. Citi was acting as the agent for a loan to bankrupt Revlon and accidentally credited the creditors accounts for the full value of the lown with the bank's own money, which started a two year legal dispute which would have been entirely avoided if Citi had just not accidentally credited the accounts.

You're complaining about Cruise. Cruise is dead. They had their license pulled in no small part due to the exact incident you're complaining about.

Now that Cruise is gone, the level of newsworthy driverless car badness has dropped from accelerating into pedestrians down to things like getting into honking matches at 4am and spending 5 minutes driving around in circles.

My understanding is that CrowdStrike has (among others) a rollout process specifically designed to address immediate active threats. This rollout is ultimately owned by CrowdStrike as part of their product.

The problem is that they didn't have a testing process even close to being commensurate with that extreme of a rollout -- "roll out everywhere now!" is incredibly dangerous, even (especially!) for updates that are only "configuration", and while they did have some automated testing those tests didn't include actually pushing the update to real machines and making sure it could be rolled back and didn't immediately crash.

I don't think there's anything inherently wrong with it being "a single day" -- there are legitimate tradeoffs with security here that want unusually fast rollouts.

Even still, if this thing had been rolled out over even a couple of hours, even with the crudest of telemetry, it would have presumably been stopped when it only crashlooped a few percent of machines. That still would have been a huge annoyance but it wouldn't have been stop-the-business bad for the large majority of affected businesses.

From what I can tell there are a remarkable number of people who appear to be grinding bullshit culture war axes rather than engaging in things like reading comprehension.

It's not that things haven't always been bad, but it feels like half the posts here the people aren't even reading the posts they're responding to.