I'm not sure if I'm interpreting the paper correctly, but I don't think those numbers quite match what the authors found. I find their presentation confusing, though, so possibly I'm the one that's wrong.
I think you understand this but in case it's not clear to others (I wasn't sure until I got to the end of the paper) all of the machines in their study are using ECC memory. This allows them to measure both the number of Correctable Errors (single bit) and Uncorrectable Errors (more than single).
I think you are right that the chances of a single bit flip for a randomly chosen machine are about 25%. In Table 1, they say CE incidence for a randomly chosen machine is 32% per year. But they also show that a machine that has one correctable memory error is much more likely to exhibit many more.
I think that "CE Median Affct" (also in Table 1) says that among the machines that have been affected by at least one correctable error, the mean number of errors is 277. Thus rather than one crash per 10 machines over 3 years, it's likely that you'll get hundreds of crashes, but with one (or zero, or two) of those 10 machines responsible for all of them.
The Uncorrectable Error rate they found was about 1/20 of the correctable rate, a little over 1% per year. Assuming the same "use time" factor applies, this would mean that with ECC none of the 10 machines would be likely to ever crash due to bit flips, reducing the number of crashes by hundreds.
But this doesn't tell the full story either. Maybe it's best to view it as 1000 employees, each of whom is issued a laptop. If you buy them machines without ECC, the initial cost is less but 100 of them will experience frequent crashes during the lifetime of the laptop, leading to frustration, possible lost work, and early machine replacement.
If you pay more up front for ECC, the number of affected employees will be reduced from 100 to 5. How much extra one should be willing to pay for this reduction depends on circumstances, but I think the impact is quite different than merely avoiding 100 equally distributed random crashes over the course of 3 years.
I agree this point is confusing. However, I think the "10% of employees will experience frequent crashes and possible data loss" sounds way too extreme.
I mean, how many companies do you know that give all their workers Xeon-powered laptops? My impression is that 99% of people have i3/5/7 powered machines at work.
Are really companies experiencing frequent failures caused by RAM on 10% of their employee's machines? Can anyone with such knowledge weigh in?
The problems with bitflip on consumer workstations is greatly overstated by some. The single biggest cause of crashes is, and always will be, the end user themselves. Or PEBKAC[1] as some like to call it.
While reducing bitflip on all devices would be a nice step forward for consumer electronics, pragmatically I have to agree with you that it would have negligible impact in the real world. Those that need it generally already pay for it, and those that don't already find countless other and more frequent ways of breaking their machines.
IIRC, most crashes are due to poor driver implementations, and other software bugs. I can't think of a user interaction that would crash a system, unless they physically beat the hell out of their system causing damage to spinning rust.
I get what you're saying, but often the bad drivers are because users are installing non-recommended drivers to begin with. Or installing known buggy software that experienced users like us would have avoided.
I've seen users place PC and Macs next to radiators or surround their device with folders and reference books, oblivious to the fact that computers can overheat and need ventilation.
I've seen users trying to use Excel as a database, then get impatience when Excel starts hanging or Windows starts running slow, and retaliate by pressing more buttons more rapidly thus locking the entire OS up.
I've seen users who would just switch the power off at the wall even night. Overtime NTFS would corrupt to a point where random crashes would happen.
I've also seen people physical attack - kick and thump - their computer when it runs slowly, an application crashes, or just through bad temper. This can damage the HDD and lead to more crashes.
There are also users who think they understand computers, so they like to tinker with lower level settings (IRQs, paging, CMOS settings, driver options, etc) then get surprised when things behave unexpectedly.
Most recently though, I had to explain to one guy who installed his own motherboard and accidentally cut a deep scratch into it, why his random crashes are potentially his own fault.
So I definitely blame users for most PC crashes. Even with the few number of kernel panics I've had over the last 15 or so years, I can attribute most of them to myself. eg I was experimenting with undocumented Windows APIs to change low level behaviors. Or I was experimenting with running non-supported file systems as my main OS drive.
I think you understand this but in case it's not clear to others (I wasn't sure until I got to the end of the paper) all of the machines in their study are using ECC memory. This allows them to measure both the number of Correctable Errors (single bit) and Uncorrectable Errors (more than single).
I think you are right that the chances of a single bit flip for a randomly chosen machine are about 25%. In Table 1, they say CE incidence for a randomly chosen machine is 32% per year. But they also show that a machine that has one correctable memory error is much more likely to exhibit many more.
I think that "CE Median Affct" (also in Table 1) says that among the machines that have been affected by at least one correctable error, the mean number of errors is 277. Thus rather than one crash per 10 machines over 3 years, it's likely that you'll get hundreds of crashes, but with one (or zero, or two) of those 10 machines responsible for all of them.
The Uncorrectable Error rate they found was about 1/20 of the correctable rate, a little over 1% per year. Assuming the same "use time" factor applies, this would mean that with ECC none of the 10 machines would be likely to ever crash due to bit flips, reducing the number of crashes by hundreds.
But this doesn't tell the full story either. Maybe it's best to view it as 1000 employees, each of whom is issued a laptop. If you buy them machines without ECC, the initial cost is less but 100 of them will experience frequent crashes during the lifetime of the laptop, leading to frustration, possible lost work, and early machine replacement.
If you pay more up front for ECC, the number of affected employees will be reduced from 100 to 5. How much extra one should be willing to pay for this reduction depends on circumstances, but I think the impact is quite different than merely avoiding 100 equally distributed random crashes over the course of 3 years.