I have been a multiple times been on a guest VM on on-premise infra where microvms were used with nested virtualization for exactly this purpose. The setups don't have a AWS Lambda to fallback too nor do they just get to have access to the hypervisor/bare metal.
This was gVisor in kvm-mode and qemu inside containers for integration tests. The main reason here is cost, it is very compelling for getting more out of a hypervisor setup.
> The graphics just don't seem complex enough to need a 5090 to render at 60fps.
I don't want to be too simplistic here bit they do seem like quite good graphics. Raytracing/modern lighting and extremely detailed models are not particularly cheap.
If you look at other triple AAA titles with similiar graphics than 4k@60fps doesn't seem unusual, exempting very optimized/older engines like cp2077.
As former HPC ops albeit in public service, we also tend to have funding for everything but ops personnel. This means you also often cannot afford the personnel you might be wishing for as they tend to go more towards better paying fields like devops etc.
The guard rails and protection for the good "AI" was mostly that it's abilities to perform actions was extremly limited.
It could not, as far as I can remember, take actions outside of passing a SSN of an endangered person to the protagonists for the initial seasons and only lived of information provided by the government surveiallance network. The latter seasons and the bad "AI" mostly changed that especially as the bad "AI" was weaponized with exactly this guardrail missing.
I would maybe tone that down that sentiment a bit and limit the adjustement.
You actually do want to be able to monitor them remotely, like most industrial/commercial refrigeration, to detect failures faster and more reliably. They should still be at least secured with normal auth/autz in addition to perimeter security, though.
If I can know when a fridge failed within the hour (even at night) then I can also dispatch a technician same/next day more easily.
Same reason website's are remote controllable. There's someone who's job it is, is to make sure the fridge is up and running, 24/7 and for the same reason that website operators/SREs don't live at each POP/DC, that fridge operator wants to remotely be able to flip some relays off and on to fix the problem without having to drive onsite, if possible. Additional, a single person/fewer people can then also be responsible for multiple/more sites.
That’s exactly right. There all all sorts of read-only architectures where the device can be monitored without every directly accessing it:
- The device’s control has a radio transmitter that emits its readings, but no receiver. Listen for those messages to monitor it but there’s not a physical path to reply to it.
- The device’s controller writes a database or message queue. Its code never reads from that source. Monitors listen to those messages.
And many, many similar methods with the same core concept: the item transmits its status but doesn’t have a method to receive replies. You just have to commit to using any one of those thousands of mechanisms.
I have a read only temperature sensor in my old-school non-computerized refrigerator and freezer. Hack until the cows come home but it still can't turn off my freezer.
Good. You've got a read-only widget that monitors your freezer.
If it were hacked to start delivering temperature readings that were dangerously-high while you were out of town, then: Would you keep the food, or discard it?
I also keep a marble on a shot glass of ice ;) but I am not good at religiously checking it every time I open the freezer. It's a sanity check for the sensor.
1) you'd wonder what broke
2) you'd wonder why it fixed itself
3) you'd check the ice cream when you got home
4) ice cream is fine, must have been a bug in the temperature reporting.
Air-gapping, network segregation, data diodes, etc. are all real things. It's possible to make a read-only refrigerator that is "unhackable". I have a temperature monitor in my refrigerator right now. Hack that monitor and it doesn't affect my refrigerator at all.
Depending on the make and model, send a reverse feed code and then overwrite with block characters. You will know that something bad happened, but you won't know what.
That makes it inconvenient to read, not impossible. E.g. use transmitted instead of reflected light (shine a bright light behind the paper) to see how much ink is in each place.
There is value in nuance and debating whether the solution should be scaled down to say a store ( or only available on local vlan or something), but I think the value of discussing that is kinda diminishing, when, apparently, the controls for all stores can be done from the same hit.
But to address your specific point:
being able to monitor does not ( or at least should not ) automatically equal put it on the web
There are most likely IT governance problems: builders bought fridges, IT didn't provision sufficient access, installers put 3G keys instead of spending time getting the correct VPN/networking, and a few months later no one knows how this thing works, and IT is not mandated to find out that shops ever did that.
This still shouldn't be online. They can be networked, if you need remote monitoring, but that's different from being online (at least in my definition).
Obviously it's a question of cost and convenience, and just dumping your appliance on the internet and make it phone home is going to be the cheaper option. If you want security those appliances goes on a separate network, which cannot reach the internet.
This line suggests the units themselves were targeted individually with a hack:
> Danfoss said the affected cold units were older models whose security software could not be updated to ward off a risk they had identified.
But the problem with networked devices spread across tens of thousands of pretty much unguarded locations, surrounded by unvetted employees, but taking orders form a central location is that there is no such thing as "not online". They'll talk to the management server used to set up the nightly defrost schedules across the fleet. Even if they are in an isolated VLAN, it's always one step away from someone connecting a Pi to that network and off they go online. There's only so much zero-trust, NAC/802.1x and segmentation can do when some devices will be too old to support these and their network will be the entry point to the management server, then to the whole fleet.
I'm honestly very surprised that a couple of things things haven't happened yet: the central server for managing these fleets getting hacked and allowing access to the entire fleet, and/or the companies realizing that the only way to limit the reach of an attack is to physically not have the hardware needed to do remote control.
So this is how it was done 25 years ago: Only the used ports (ethernet jacks) on a network is activate, disconnecting a device and the port closes. That strictly speaking doesn't mean that there aren't way to get devices on the network, but it does make it a lot harder. You're absolutely correct about these networks being in hostile environments, but why then do we pretend that they are not? Fridges like these have never not been in a hostile environment, so why aren't basic security measures taken to secure the physical side of things?
In my world the central server can be hack in two ways, that is by a malicious operator or via some scheme like the US/Israeli attack on Irans uranium centrifuges. The central server also have no business being online, that to is on a closed network, running the whitelisted applications from the manufacturer.
This is how it was done in this [0] very nice 60 years old (I'm sad now) movie. [Spoiler alert] An alarm is triggered so many times the guards just disable it. Alert fatigue has worked the same forever. Nobody wants to check the thousands of false alarms that probably happen every day across thousands of locations.
As long a schedule is conceived in one place and applied in another, even with human intervention on both sides, either one of the sides can be swindled. At this point with the bar for a successful hack being lower than ever, completely removing the hardware necessary to make remote changes is probably the only way to limit the blast radius.
Maybe allowing to send the schedule but requiring a local employee to operate a local hardware only control to apply it could work, but who has the money for this kind of complexity. Something like this on every freezer (device, really) in a fleet probably costs multiples of the money they lost with the spoiled food.
this definitely does not require a fridge be "online" and have the ability to be remotely programmed. At the very least have the monitoring and potentially I/O capabilities in a different device that can easily be removed from the system. Integrated smart fridges deserve to Suck it, Jin-yang!
> we can't assume present returns will match future returns.
Funnily enough, after spending some time in public service albeit in another country, genereally budgeting for public agencies works almost inverse to this.
If you get allocated X budget but only spend 0.8 * X, then next time you will almost always get
Its a very difficult problem. The exact same dynamic happened in the USSR (my Dad has stories that are basically word for word what is described). I have also seen it in large corporations and US state and federal governments. I would not be surprised if we had papyrii from ancient Egypt about the need to spend the pyramid budget this year or the pharaoh may not allocate enough next year.
Sure but almost no OSS projects build containers for Windows (to my knowledge). The same goes for XNU/darwin afaik. They may get binaries but container images often assume Linux by default.
Developers, ime, also don't really want to start building their own container images for auxiliary services like, e.g., PostgreSQL, MongoDB or MySQL.
Having virtualized Linux support means you can take the existing ecosystem and make it accessible to the other two.
It also means docker itself doesn't have to fully embrace & support either which is part of why it's "cheaper".
They do take it mostly into account for IC(E) journeys and some regional trains.
You will usually have an announcement like "Alle Anschlusszüge werden weiter erreicht werden" oder "Zug nach $ort wartet auf uns" here.
It just doesn't propagate to traffic within a federal regional (Bundesland) a lot which sucks but is somewhat understandable (the frequency there is supposedly high enough although connected journeys kinda break this).
it writes to the file with elevated priviledges and doesn't write it back to the terminal, effectively a 'sudo sh -c '{cmd} > {file}''.
sudo tee is a common way to work around priviledge restrictions in shell pipes (you can't easily 'sudo' a pipe), if /secure was 0600 and owned by root:
"sudo echo "test" > /secure" or "sudo printf '' > /secure", would not work because your user's shell process can't write to /secure. However, "echo 'test' | sudo tee /secure" works because the elevated 'sudo tee' can write to the file.
This was gVisor in kvm-mode and qemu inside containers for integration tests. The main reason here is cost, it is very compelling for getting more out of a hypervisor setup.
reply