On Tue, 2019-05-21 at 19:39 -0500, Eric Blake wrote: > On 5/21/19 6:52 PM, Aarushi Mehta wrote: > > Signed-off-by: Aarushi Mehta <mehta.aar...@gmail.com> > > Sparse on the details. The subject line says what, but without a > 'why' > for how io_uring is different from existing aio options, it's hard to > see why I'd want to use it. Do you have any benchmark numbers?
For peak performance, io_uring helps us get to 1.7M 4k IOPS with polling. aio reaches a performance cliff much lower than that, at 608K. If we disable polling, io_uring is able to drive about 1.2M IOPS for the (otherwise) same test case. More details, and the source for the above is at http://kernel.dk/io_uring.pdf > > --- > > qapi/block-core.json | 3 ++- > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > diff --git a/qapi/block-core.json b/qapi/block-core.json > > index 7ccbfff9d0..116995810a 100644 > > --- a/qapi/block-core.json > > +++ b/qapi/block-core.json > > @@ -2776,11 +2776,12 @@ > > # > > # @threads: Use qemu's thread pool > > # @native: Use native AIO backend (only Linux and Windows) > > +# @io_uring: Use linux io_uring > > Missing a '(since 4.1)' tag. > > > # > > # Since: 2.9 > > ## > > { 'enum': 'BlockdevAioOptions', > > - 'data': [ 'threads', 'native' ] } > > + 'data': [ 'threads', 'native','io_uring' ] } > > Missing space after ',' (not essential, but matching style is nice). > Should the new element be defined conditionally, so that > introspection > only sees the new enum member when compiled for Linux? > I'm not sure what would be the benefits of that? We already check for Linux at configure, and this would reduce readability. We aren't doing this for native. >
