I want to put an example.

If I have a motherboard, some elements dropped (such as capacitance or resistance). When I install it into a computer, it can pass Power on Self Test and successfully launch the OS. Then I can operate it as normal.

In this situation, can I think the device runs normally? Can I believe its operation result (calculate result or logical result )?

  • \$\begingroup\$ The POST is not exhaustive. The hardware might boot and work well for 99.999% of the time, but then fail when you least expect it, and more worryingly, not realise that it has failed. In short, the answers are yes and maybe respectively. Sort of related, google "Muntzing"... \$\endgroup\$ – Greenonline Nov 9 '19 at 7:15
  • \$\begingroup\$ @Greenonline Hi, thanks for your comment. Yes, I also think the worst thing is byzantine errors, as you said. But if the system makes miscalculations or logical errors, won't cause the system to crash? \$\endgroup\$ – Jeason Nov 9 '19 at 7:54

Yes, of course hardware damage can be so subtle that it does not cause catastrophic failure and smoke right away, but just increases probability for a bit to flip here and there. When a single bit flips, it matters a lot where this happens, causing just wrong results or complete crash of software or operating system.

Having said that, even if the device is in perfectly good condition, cosmic rays or radioactive decay can still flip some bits in memory so it can still operate wrong or crash the whole system.

  • \$\begingroup\$ Indeed, and this is why for safety-critical computerized control systems such as in airplanes, a lot of effort is put into detecting/avoiding such errors, for example by redundancy. \$\endgroup\$ – Erlkoenig Nov 9 '19 at 9:56
  • \$\begingroup\$ "of course"? The presence of the question would suggest this is not so obvious to all. What leads to the increase in the probably of a bit flip? Is it related to the well-known (random) causes such as those you mention? Are there any references on this phenomena? I don't recall this cause being mentioned in the TÜV Functional Safety course, which is body of work that guides safety-critical systems. Redundancy seeks to mitigate against failures and random errors. The calculations assume failure rates increase over time (the bathtub curve) but random error rates are constant. Is this not true? \$\endgroup\$ – Heath Raftery Nov 9 '19 at 11:04
  • \$\begingroup\$ The original post talks about standard consumer PC motherboard or smartphone, not super safety critical system. Regardless of that, such device is manufactured to operate over a given range of specifications such as temperature range, voltage range, voltage ripple range, etc. Also the electronic components such as capacitors, resistors, etc, are manufactured to operate over a given range of specifications. If the motherboard has a part e.g. capacitor that it is accidentally poked off with a screwdriver, you change the device and it can then only work within smaller tolerances, if at all. \$\endgroup\$ – Justme Nov 9 '19 at 12:07
  • \$\begingroup\$ As @Justme said, in this post, we only discuss about consumer pc or smartphone. In my opinion, when the OS launched, the pc or the smartphone made a lot of calculation or logic operation. If the bits filp was much enough, cause too many wrong results, the OS cannot be launched. In other words, once the OS launched and can operate it as normal, it's prove that the system don't have too serious problem, so I can trust its operation results. Am I right ? \$\endgroup\$ – Jeason Nov 9 '19 at 14:58
  • \$\begingroup\$ There is no way to guess if you can trust the system or not. If some component is missing or malfunctioning, the system might continue operating normally in every situation. Or it may work while it is at room temperature, but it might fail when put under heavy load and it uses a lot of power and it heats up, and thermal expansion bends components slightly. So it is really up to you in what way and how long you want to test it and determine if you trust it enough. \$\endgroup\$ – Justme Nov 9 '19 at 15:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.