LINDSAY@TL-20A.ARPA (08/08/84)
Calculations based on failure rates are very useful, but they gloss over the fact that the rates are for good chips, that haven't been stressed, that aren't wrought into real systems, and that haven't aged. When you manufacture (say) a million PCs, you get infant mortality that evades your system-level burn-in. Repair people will cause brief overvoltages, or will overheat components while soldering. Units will be exposed to ocean breezes (salt spray), and people will buy dubious add-ons that e.g. drag down the power rail, and they will get cigar ash in the sockets while populating a perfectly good ram board. Batches of chips will be used and shipped before it is noticed that they have low drive at high temperatures. Then, there's Vonada's Law (the worst cases will only add up in the best customer's machine). Some fool will keep one in his sauna. The production line will have off days, and there WILL be lemons. Users will plug the printer into a different wall, and cause ground currents. (Not to mention booting the wrong disk, but that's a different bulletin board.) The point I am driving at is that real-system failures tend to cluster in certain machines, and the more popular your product, the more this will be so. (Success as a danger! ) Preventive measures (such as parity) should be aimed at keeping the owners of those machines productive, if only by getting them to a repairman. -------