On Sun, Jan 31, 2021 at 11:57 AM Matthew Miller <mattdm@xxxxxxxxxxxxxxxxx> wrote: > > On my Intel i7 laptop, xxhash is a small but clear performance win over > crc32c: > > $ ./hash-speedtest 10000000 > Block size: 4096 > Iterations: 10000000 > Implementation: builtin > > NULL-NOP: cycles: 1372543560, c/i 137 > NULL-MEMCPY: cycles: 2844174884, c/i 284 > CRC32C: cycles: 9673117404, c/i 967 > XXHASH: cycles: 7129819594, c/i 712 > SHA256: cycles: 649914613520, c/i 64991 > BLAKE2b: cycles: 153513008046, c/i 15351 > > > And I'm given to understand that this is even more the case on newer CPUs. > > Plus, it's 64 bit instead of 32 bit. The 256-bit algorithms are obviously > much, much slower and probably not right for a default, but should we > consider making xxhash the default for Fedora Linux systems with btrfs? I'm seeing: [ 1.179693] Btrfs loaded, crc32c=crc32c-generic, zoned=yes Someone adventurous might want to figure out how to get it to use crc32c_intel instead. I think it might need to be compiled into the kernel. CONFIG_CRYPTO_CRC32C=y CONFIG_CRYPTO_CRC32C_INTEL=m And then retest. But also, the difference in performance between xxhash and crc32c is pretty tiny. I'm using all four hashes on various file systems, and up until there's a heavy IO workload, there's no performance difference among them. -- Chris Murphy _______________________________________________ devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/ List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/devel@xxxxxxxxxxxxxxxxxxxxxxx
