I think you are assuming microcontrollers are only for professional use and are ignoring the hobbyist market (a big part of Adafruit's market). There are a lot of people who don't need to care about the most efficient, reliable, or cost effective use of resources. I'm one of those people and use CircuitPython/MicroPython for my simple, one-off needs all the time.
While I tend to think local management is preferable, some of the NWS services might actually benefit from a cloud deployment. The benefit/risk/cost analysis sounds pretty involved:
- The NWS services are mission critical. Hosting needs a high level of redundancy and distributed alternate sites. The large commercial cloud providers can easily provide widely distributed services with automated failover routing and built-in distributed backups.
- The load level of the public-facing services are not consistent. During severe weather events demand spikes, possilby 10x or more (I wouldn't be surprised to hear it was even higher than 10x). You either have to provision servers and bandwidth for the surge levels or design a system that will degrade gracefully and limit data services without too much inconvenience to the customer. Large cloud providers can provide on-demand server and network scaling. The alternative is paying for a lot of idle equipment and unused network capacity at (hopefully multiple redundant) locally owned sites.
(Note: Yes, If you can invest in multiple distributed sites, you can reduce the redundancy by spreading out your services among all sites during normal operations, but you still have to plan for the contingency of one location having to serve all the traffic).
I can't even guess what those numbers are for NWS, but I suspect that at least some services would be better off with commercial hosting. I also suspect that the "right" balance would be a hybrid architecture.
Do I think a balanced solution is likely to happen? No way. Its a government agency... it would take a superhuman level of effort to keep the higher-level administrators from throwing the baby out with the bathwater. Just hinting that you could avoid having to deal with owning IT equipment is enough to bring tears of joy to any budget analyst and property manager. After all, once its done no matter how much the cloud services cost it requires less work than trying to keep track of a million pieces of equipment and a thousand service agreements.
I'm watching this train wreck in motion in a few projects. All I can do is try to convince them they need to have an exit strategy that will let them keep control of their data once the customer decides they don't feel like paying for services anymore.
I agree about bandwidth planning, especially for the radar site (formerly the most valuable weather-related site I used). I recall a couple of times over the last few years where they put up an interactive map prototype for comment. Those prototypes were dreadfully slow, even when considering that it was probably being served from an overloaded test server. I suspect the "final justification" for the current mess was the loss of flash for the "enhanced" display (on a couple of surveys, I'd suggested they either just make the "standard" animated GIF display the default again, or try to rework parts of the flash design into a fairly simple javascript layer that presented the same GIF data). Those comments were submitted using the customer surveys, though, so I doubt the text comments ever made it to anyone to actually look at.
Instead, we get the current monstrosity. Granted its not a really terrible design (mostly, there are still some bad parts and missing features like controlling the history display), but the tile-based layer services are a horrible choice for critical high-load data. While its nice to say you are using web GIS standards, if you can't actually service the requests, you become worse than useless (does the clear area mean there isn't anything to report, or did you just not happen to be able to serve tiles for that zoom level?).
Statement of Bias: Pinephone (bravehart version) and PineTab owner. Pleased with both, but not using pinephone yet.
There are reasons for the price differences, and there are very good reasons for both approaches.
The Librem phone is the product of a for-profit company. They have a strong incentive to build a "finished" and functional product with working hardware and fullly integrated software (maybe evertyhing doesn't work now, but as long as there is a product to sell, they will be working to get there). I think the rule of thumb is that a reasonable final price for a product is something like 4x the cost of the bill of materials; I don't think this includes software-related costs. For a higher cost, you get a finished product out of the box with one point of contact for support. This is the tried-and-true model where the manufacturer stands behind the product. The difference here is that normally the producer tries to tightly control the software stack (if only to minimize support costs), but Puri.sm doesn't try to control the software.
IIRC, Pine64 is a non-profit (not sure how that works) that designs and releases the hardware for not much more than the cost of production/storage/shipping (most salaries can easily be put under those expenses, so you can still have a non-profit structure). Software support is minimal, the community is expected to actually make the product into something as it sees fit. As shipped, the original pinephone version just booted into a self-test mode. The rest was up to the user to set up, with multiple environments to choose from (none of which could even reliably make a call when the "braveheart" pre-release version was shipped). This model has the advantages of getting low-cost hardware out to as many people as possible to foster innovation, but you end up with a fragmented user base. My phone's speaker isn't working... it is the hardware, or just the version of postmarketOS I'm running?. There is no one-person I can ask, its going to take a while to troubleshoot. On the other hand, I have lots of different groups with lots of different ideas of how to make a phone work. Natural Selection may eventually weed out some of them, or I may choose to a niche system that exists along with other oddballs on some galapagos island of the internet. This is a lot like the "white box" PC world that led to the flourishing of Linux in the first place. I think the future of linux-based phone environments will come more from this approach than the other, but it will be a longer, slower, and more "painful" route.
Both approaches are Good and Right.
If anyone is looking for a the definitive collection of Leher's work, may I suggest "The Remains of Tom Lehrer"
https://en.wikipedia.org/wiki/The_Remains_of_Tom_Lehrer
My wife bought me a copy as a gift many years ago (...I Love That Woman...). I was shocked when I looked up what its going for today.
If nothing else, the track listing can be used to help decide what individual albums and tracks are necessary to get a complete set of his works.
As a chid of the 70s, the Electric Company tracks really make me nostalgic. I was sort of creeped-out by "Silent-E" (the thought of turning a "twin" into "twine", or a "dam" into a "dame", not to mention the trippy E-with magic wand image... but at least "my friend Sam, stayed just the Same").
"Wernher Von Braun" should be the official anthem of Redstone Arsenal. "Vonce the rockets go up, who cares vhere ze come down, 'Zat's not my department' says Wernher Von Braun"
Maybe Goku found them all?
There still appear to be a lot of PICMG and PCISA boards/backplanes out there.
Its been a while since I dealt with industrial passive backplanes, but they at least used to be a reasonably affordable solution when either old ISA cards, or large numbers of bridged PCI slots were needed. Not cheap, but reasonable, especially when you consider the entire market is built around industrial-grade requirements and the expense of moving to newer controllers.
Went on e-bay to look at used equipment and got a little nostalgic (I'm looking at you Pentium II card...).
I failed until too late to learn one of the biggest lessons of college: Some things are not worth the effort.
This was over 30 years ago, so things may have changed, but my biggest time sink wasn't the 100 level classes, it was the Labs. I would spend as much time (sometimes more) working on lab reports for a 1 hour class than I spent studying the material for the 3 hour main course. Granted, the lab itself was fun, but in the late 80's it took a lot of effort to do the reports that were formatted well enough to not get an automatic B (or C) (I could not afford my own PC until senior year, by then the labs were over).
I later realized how badly I'd misused my time. I should have accepted lower scores for the labs and focused on the real coursework. If I'd just done the work, did a barely passable report, and spent the time saved on more important things, I think I'd have been better off.
I've come to believe that the "trade-space analysis" was one of the lessons they were really trying to teach. In retrospect, it may have been hinted to many times by the staff, I just completely missed the point.
Reminds me of a story a fried told me. He'd bought a Quest to allow him to watch movies while laying in bed so he wouldn't disturb his wife at night. He set up the device and decided he liked a log cabin setting where you were positioned on a couch looking at a large screen TV.
The first night he tried to use it, he got in bed and started the movie. Then he laid back on the pillow and headset tracking faithfully followed his head motion. If he didn't prop himself up, he would end up just staring at the ceiling of the cabin environment while the movie played on the virtual TV out of view.
I don't know any details about what services they run with what provider, but consider:
- The article says they already do business with Amazon and Microsoft.
- Now they are adding a third provider (Oracle).
As long as they do not fall into the trap of becoming reliant on any particular vendor's infrastructure, they should be in an excellent position to negotiate when they end up scaling back operations.
"Oh, by the way, since out loads have dropped, we'll be ending our relationship with one of you. We already know how to operate in all three environments, so cost is going to be the deciding factor. Would love to keep a relationship with all of you, but we simply can't afford that. Sorry."
I wouldn't be surprised to see them add Google at some point to increase the number of potential hosts.
I suspect a collaboration site is mainly connection juggling and fairly straightforward back-end data storage ("straightforward", not "simple" or "easy"). While its probably a nightmare to manage and scale, a good design should probably be very portable between different infrastructure vendors. Easily move the basic code in place and then start tuning to fit whoever is hosting you.
Long-term data management and migration is another issue. But as long as those considerations are in place from the start, it becomes a solved problem for each vendor.
Your code should be more efficient!
