On Mon, Aug 18, 2008 at 04:41:35PM -0700, paul wrote: > CORRECTION: As fletcher4 actually sums 32bit (not 64bit) data using 64bit > accumulators, > the first two checksum terms (a and b) will not overflow for data block sizes > as large as > 128KB, and therefore should be considered at least as strong as that expected > of a 32/64bit > Fletcher checksum (although the remaining terms (c and d) may overflow); and > thereby have > a worst case hamming distance of at least 3 for zfs's maximum 128KB block > size; which is a > good thing. (sorry for my earlier miss-diagnosis) > > fletcher2 (zfs's default) however remains arguably flawed, however may be > easily strengthened > by correspondingly being restricted to summing 32bit data using 64bit > accumulators, or possibly > alternatively summing 64bit data using 128bit accumulators if it's viewed > reasonably supportable > by target architectures/compilers likely to host zfs.
Thanks for the report. I've filed: 6740597 zfs fletcher-2 is losing its carries to track this. Cheers, - jonathan > -- > This messages posted from opensolaris.org > _______________________________________________ > zfs-code mailing list > zfs-code at opensolaris.org > http://mail.opensolaris.org/mailman/listinfo/zfs-code
