Fragile non-ecc systems?

Something that works absolutely fine 99.9999% of the time is not fragile.

Probability of a bit error increases as memory increases. Opens a large can of worms if control flow breaks.

Yep. 99.9999% correct would mean 8 bit flips per megabyte of data stored in ram. The error rates are (thankfully) much lower than that (otherwise your computer wouldn’t boot). But random bit flips can cause utter havoc if they happen at the wrong time or place. If you download software from the internet, bit flips can introduce weird bugs to your software, only on your computer. (Including in the OS - including your filesystem or drivers). They can corrupt writes to your hard drive, and as a result corrupt your drive or files. Bit flips can quietly change the DNS request your browser sends to cause terrifying security problems. Or edit forms before you send them. There was even a case of a voting machine in Germany accidentally inventing 4096 votes due to a bit flip.

ECC is a really good idea. It’s only expensive right now because it’s a “premium feature”. If it were a standard part of all ram sticks, it’d be cheap and we’d all benefit.

Not argueing against ECC, but some of your scenarios seem to be outdated due to crypto. I.e software you download from the internet is often signed and hashes are validated (Linux package managers, macOS developer certs). Same for DNS requests (dnssec) etc. Yes, there is still wiggle room for bitflips to cause problems, but less so than in the past.

DNSSEC is around the 4~5% mark in .com and .net.

True - my bad of referring to DNSsec; there are other ways you can use encryption for DNS resolving (by using an external DNS server that encrypts using TLS or simply by using DNS-over-HTTPs). This way you get 100% encryption of your DNS traffic (and thus tamper checks that would detect bitflips). Again, not arguing against ECC, there are valid points to want it - I just see less and less reasons in the consumer market.

Encryption and signing don't protect against memory corruption.

For example, I download software from the internet then hash it. The hash matches. Before the bytes are written to disk locally, a bit flips in RAM. The corrupted data is written to disk and used.

Likewise, dnssec doesn't protect you against DNS bitsquatting attacks[1] because the domain name can be changed before the DNS request is made. So the DNS response your computer makes for a-azon.com might be totally valid and signed. It can come through DoH or whatever. The problem is that your browser thought it was the response for amazon.com and chrome send a bitsquatter your amazon cookies. (Oops).

[1] https://www.youtube.com/watch?v=9WcHsT97suU

Well aware of all of that, but it decreases the chances if corruption (ie no corruption during download).

news.ycombinator.com not included in this 4% either

That's just made up statistics - average computer gets multiple errors a day.

This was fine when consumer computers were for games and porn. Now they store birth certificates, submit information to the tax man, documents to court, sometimes deal with matters of life and death

>legislators

Haven't you seen those congressional hearings with Zuck and others where congresspeople ask about how to use their phone ? How do you expect them to legislate about ECC ?

I hear this all the time, that non-ecc machines are just completely fragile, and the math seems to indicate that one should be getting bit flips all the time. and yet, my Intel system has 64gb of non-ecc ram and it runs half the day everyday and is hibernated at night. I run 3d cad software on it, Photoshop, vs code with tons of extensions and software running in wsl2 docker containers and I just never really get any errors. certainly no crashing or blue screen. what exactly should one expect from bitflip errors? I would think that even single bitflips as often as they should happen with 64gb of ram being used almost fully would show up somehow.

> what exactly should one expect from bitflip errors?

Probably silently wrong values happening in memory somewhere. If it's not in something where that value is not actively running code code, you probably won't have a crash.

If it's in (say) Excel calculation formula it'll probably just screw up the calculation. Which may or may not be obvious. Similar thing for 3D cad, it could be completely non-obvious.