On Sat, Dec 08, 2007 at 01:42:29 +0000, Alan Cox <alan@xxxxxxxxxxxxxxxxxxx> wrote: > > If you are doing complex scientific processing or banking where an error > isn't acceptable then its essential. If you can tolerate the small change > of a bit flip ever year or two then its far less of a concern. I sat in on a discission with the group doing analysis of LIGO data at our University and while they were doing scientific work, they concluded that they didn't need to use ECC memory because the odds of a false negative were acceptibly low and for false positives they could just rerun the test. This is from the same people that found the error correction in TCP/IP not to be sufficent and worried about how to check the files and to be able to reacquire reason parts of the files as they were large enough that resending whole files after a checksum failure was not ideal. So just because the use is scientific, buying ECC memory may not be the most economical solution to the problem. P.S. Sorry about the duplicates. I forgot to delete the message from my outbound email queue after waiting for the first copy to be sent. I still haven't heard back on my infrastructure ticket complaning about getting deferral status back for messages that seemed to be getting accepted for delivery to redhat's list server for the last 2 weeks.
