On 3/4/25 00:34, Caleb Sander Mateos wrote:
On Mon, Mar 3, 2025 at 7:51 AM Pavel Begunkov <asml.silence@xxxxxxxxx> wrote:
Add registered buffer support for vectored io_uring operations. That
allows to pass an iovec, all entries of which must belong to and
point into the same registered buffer specified by sqe->buf_index.
The series covers zerocopy sendmsg and reads / writes. Reads and
writes are implemented as new opcodes, while zerocopy sendmsg
reuses IORING_RECVSEND_FIXED_BUF for the api.
Results are aligned to what one would expect from registered buffers:
t/io_uring + nullblk, single segment 16K:
34 -> 46 GiB/s
examples/send-zerocopy.c default send size (64KB):
82558 -> 123855 MB/s
Thanks for implementing this, it's great to be able to combine these 2
optimizations! Though I suspect many applications will want to perform
vectorized I/O using iovecs that come from different registered
buffers (e.g. separate header and data allocations). Perhaps a future
improvement could allow a list of buffer indices to be specified.
That's the design decision made, otherwise the API is becoming a mess
as well as handling. The user has to be smart and keep a small number
of large registered buffers and potentially growing them, which is a
good thing regardless of this feature.
--
Pavel Begunkov
