NW, ECC means error checking and correction. We've known for years that the cells and latches that store information in computer memories and registers (especially memory) are susceptible to "hits" from things like Alpha and other subatomic particles. If these particles hit a cell storing a "1" or a "0" with enough energy, they can flip the cell to the other state, which makes it wrong, of course. So, years ago, IBM and others designed error checking and correction logic into the control section of their computers such that the ECC logic could tell when a bit had gone bad. It would then correct it before passing it on the the CPU.
They first put ECC on main memory in the 70s because it was the most susceptible to Alpha hits. Then, in the 80s, IBM and others, I'm sure including Tandem, put ECC on cache memories also, because the technology for that also had become so micro-sized that Alphas could flip them too. Intel based, like Compaq Proliant servers, got cache ECC in about 1997. Turns out that the mysterious outages Sun was experiencing last year and this, was finally admitted by Sun to be because of no cache ECC on even their top of the line servers. Bad, bad oversight, especially when they tout their reliability so highly.
Tony |
