Lance: Thats correct, but the purpose of the journal is to roll back any failed writes to the filesystem. A memory error and resultant system crash would result in a corrupted file with or without data ECC. The journal protects us against that by rolling back the failed write.....
You're assuming that a memory error will always result in a system crash.
A memory error can cause a value anywhere in memory to become incorrect. In some cases this will cause no harm at all, e.g. if the memory isn't in use. In other cases it might cause a single pixel in the image being edited to become the wrong colour. It might cause a single application to crash. It might corrupt data that's about to be written to the disk - depending on the file format and where the error occurs this could result in anything from a single incorrect pixel to an unreadable file. It might corrupt part of the file system. In some cases it will indeed crash the system.
I'm not in any position to say how serious the risk of memory errors is, but for what it's worth here are a few points to consider:
Microsoft recommended ECC memory for Windows Vista, because they found that memory errors were now a major cause of Windows crashes:
http://www.tgdaily.com/hardware-features/24190-microsoft-to-encourage-use-of-ecc-memory-for-vistaThis shows that Microsoft at least believed that memory errors are worth worrying about. Unfortunately the hardware manufacturers ignored them, because ECC costs money. With the amount of memory in computers increasing, I doubt that the risk of memory errors in decreasing.
Note that the only reliable data that we have on error rates is from ECC memory, because non-ECC memory has no way to detect errors. It's probably safe to assume that ECC memory is mostly good quality, since the whole point of ECC is reliability. With non-ECC memory the error rates may be a lot higher, at least for the lower quality modules. Back in the days when memory was a lot more expensive and my budget limited (1995 and 2002), I had two bad experiences with defective no-name memory modules. In both cases the systems would crash many times per day. After I replaced the dodgy memory the machines were fine.
What worries me is that there may be a lot of memory out there with high enough error rates to cause a lot of mischief (randomly corrupting files, occasionally crashing programs, etc.) but not high enough for you to identify the problem and replace them. Google's study indicated that error rates varied widely between modules, even for the (presumably good quality) ECC memory.
As I've said before I may well be worrying about a non-issue, but I prefer to play it safe.
