On Tue, Jun 14, 2022 at 03:39:40PM +0100, Richard W.M. Jones wrote:
On Tue, Jun 14, 2022 at 07:58:08AM -0500, Eric Blake wrote:
> Add new methods nbd.Buffer.{to,from}_buffer that avoid the copying
> present in the existing nbd.Buffer.{to,from}_bytearray. The reduction
> in copies possible by this approach is no longer quite as necessary,
> now that aio_p{read,write} have been converted to take buffers
> directly. However, there is still one (marginal) benefit: if you use
> h.set_pread_initialize(False), and create a new buffer for every I/O
> (rather than reusing a buffer pool), nbd.Buffer(n) handed to
> h.aio_pread then buf.to_buffer() is slightly faster than handing a
> bytearray(n) directly to h.aio_pread, because we are able to skip the
> step of pre-zeroing the buffer.
> ---
>
Acked-by: Richard W.M. Jones <rjones(a)redhat.com>
Rich.
Now in as commit 5b74b04
--
Eric Blake, Principal Software Engineer
Red Hat, Inc. +1-919-301-3266
Virtualization:
qemu.org |
libvirt.org