mbox series

[RFC,0/2] optimise split bio

Message ID cover.1609875589.git.asml.silence@gmail.com (mailing list archive)
Headers show
Series optimise split bio | expand

Message

Pavel Begunkov Jan. 5, 2021, 7:43 p.m. UTC
With the no-copy bvec patches and nullb eats blk_bio_segment_split()
eats >7%. It adds yet another fast path for it.

        8K              16K             32K             64K
before: 932             904             868             788
after:  934             919             902             862

Would appreciate if anyone knows off the bat typical queue_max_segments, etc.
numbers for NVMe.

Pavel Begunkov (2):
  block: add a function for *segment_split fast path
  block: add a fast path for seg split of large bio

 block/blk-merge.c | 107 +++++++++++++++++++++++++++++-----------------
 1 file changed, 68 insertions(+), 39 deletions(-)