Back in the late 70's our latest model of nuclear safety monitoring system used multiple minicomputers for redundancy. For coordination, they used a type of memory that could be shared among several CPUs. We wrote all the code according to the manuals, but when we tested it, it didn't work right. The systems guys went over it and over it; we spent thousands of man-hours chasing the problem.

We swapped in several different memory units to make sure it wasn't a hardware problem.

Turns out, the hardware was misdesigned, and just plain did not work. The vendor failed to inform us that this hardware had never been sold to anyone, and never tested.

We were dumbfounded. In that era, it was unheard of for an industrial computer vendor to deliver hardware that had not been thoroughly tested. It never crossed our mind that the hardware had never before been used, and the software manuals were basically a tech writer's conjecture.

That's a mistake no engineer would make today. Times certainly have changed. Industrial computer hardware then was above suspicion; now we assume hardware and software have something wrong somewhere. The snafu was a harbinger of mounting pressure on the minicomputer industry; within a decade most of these highly touted companies were gone.

John P., Dexter, Mich.