jerryc41 wrote:
Interesting. About forty years ago, a friend asked me if a cosmic particle could change something on a disk, changing a Yes to a No. I thought that with all the data it takes to produce software, changing a bit or two wouldn't do something like that. I could see it causing a glitch, but not letting the program run and produce wrong results. I'm surprised about the Livermore report.
You’re right Jerry, an occasional flipped bit often does no harm, but at Livermore, they would run models/simulations sometimes for days at a time, and since it was on a VERY large machine with a ton of storage and zillions of disk accesses, that one bit here and there would accumulate to skew the model over time. As I mentioned, it didn’t flip the bit(s) on disk - when we read the data on the disk, it was correct - it would flip bits in the read cache on the drives, which is RAM.