Revised AVX-512 xor_gen() Implementation For Linux RAID Yielding More Performance Gains
A few days back I wrote about Google's Eric Biggers spearheading an AVX-512 implementation of xor_gen() as the Linux kernel function used for generating and validating parity blocks such as for RAID5/RAID6. That initial implementation was yielding up to 41% better performance while a new implementation has now been posted for scoring some additional victories.
Biggers has been working through an AVX-512 version of xor_gen() to help with Linux software RAID performance and the like, that function is also used by some Linux file-systems directly like Btrfs too.
With the new implementation posted overnight, it's now up to a 43% improvement in performance while other src count sizes are benefiting more than the original implementation:
That v2 implementation is now out for review on the Linux kernel mailing list.
Biggers has been working through an AVX-512 version of xor_gen() to help with Linux software RAID performance and the like, that function is also used by some Linux file-systems directly like Btrfs too.
With the new implementation posted overnight, it's now up to a 43% improvement in performance while other src count sizes are benefiting more than the original implementation:
That v2 implementation is now out for review on the Linux kernel mailing list.