Displaying 1 result from an estimated 1 matches for "1x16k".
Did you mean:
116k
2010 Nov 02
2
[RFC][PATCH] direct-io: btrfs: avoid splitting dio requests for non-btrfs filesystems
...y the test for:
"dio->final_block_in_bio != dio->cur_page_block".
As this was in the code for a long time I just assume it is right.
So eventually for the 64k request in the example this patch improves the
effective submissions that get to the block device layer from:
10x4k, 1x8k, 1x16k to 1x48k & 1x16k which is much better.
Througput impact is small, but in terms of cpu consumption this is visible
by a single digit percentage depending on the incoming request size.
The solution looking for comments or alternatives in this RFC patch adds a new
kiocb flag that let filesystems...