📜 ⬆️ ⬇️

Empirical analysis of hardware failures on a million PCs

Microsoft conducted the first ever large-scale study of hardware failures on millions of personal computers ( PDF ). Revealed some interesting facts.

In contrast to the common opinion about the triviality of hardware failures, in fact they are quite rare, with 99% of failures being repeated. For example, a machine with accumulated 30+ days of CPU operation over a period of 8 months has the probability of failure of 1/190 due to an error in the CPU subsystem. If this happens, the probability of re-failure on this machine is 1 / 2.9.


')
If the first failure occurred within five days of the CPU, then 84% of these computers show a repeated failure within 10 days, and 97% within a month.

Overclocking CPU increases the probability of failure by 4-19 times, depending on the brand of processor.



The probability of DRAM failure during overclocking increases five times.



On the other hand, the processor at a lower frequency increases the reliability of the equipment.



As part of the study, the probability of CPU, DRAM and disk subsystem failures was calculated, on desktops and laptops, on computers of well-known brands and self-assembly. The dependence of the number of failures on the age of the computer, the amount of memory, CPU performance is shown.



Another interesting fact: it turns out, laptops usually work more reliably than desktop computers.



Methodology
The analysis was carried out in 2008 on the basis of crash / status reports sent by the Windows Error Reporting system in cases of usual failures or after a reboot of Windows (respectively, two samples: RAC and ATLAS). In crash reports specified period of work without failure. The study is considered conservative because it does not take into account cases where crash reports are not sent due to a too high failure rate.

Information collected from 950 thousand computers.

Source: https://habr.com/ru/post/146608/


All Articles