Bill Bogstad wrote: > ECC is not 100%. It's not intended to be 100%. It's intended to fault when cosmic ray strikes causes random bit flips.
> Nor does it protect against transient CPU/memory > cache errors during > checksum computation. If you are saying that ZFS can then I will happily read You have a block of data in memory, and you calculate a checksum, and write the data and checksum out, and the controller says that the writes are completed, and you read that checksum and data back and calculate a checksum on the data that was read. Then compare the checksums. If they match? What's on disk matches what's in RAM. If they don't then there's a problem somewhere and the file system driver knows it. There's more to it. ZFS and Btrfs don't just checksum blocks and extents. They checksum the checksums. There's a checksum hierarchy from the root on down (Merkle tree). Assuming that you have ECC RAM and disks that honor sync commands then yes, ZFS and Btrfs are doing some kind of awesome and you really should read the whitepapers about them. -- Rich P. _______________________________________________ Discuss mailing list [email protected] http://lists.blu.org/mailman/listinfo/discuss
