Hello!
On Sat, Feb 27, 2021 at 12:48:04PM +0000, Zhao, Ping wrote:
> Yes, io_uring can help Nginx achieve same performance with lower
> resource cost. This is the key improvement of io_uring I think.
> It can't break through the HW limitation.
Thanks for the patch and testing.
Potentially this looks interesting, despite the fact that
improvements even in terms of resource costs seem to be minor
compared to properly configured nginx using other I/O variants.
There are, however, some concerns regarding the interface itself
and tunings need to be applied in order for it to work, as well
as regressions in various kernels out there. There are also some
questions about the patch, notably retries in case of short reads,
though this probably needs better understanding of the interface.
As of now, the consensus is that we'll get back to this some time
later, though probably we want to make this not a default aio
method at first, but rather an alternative one, available with
something like "aio io_uring;", similarly to "aio threads;".
Thanks again for your work. We'll contact you once we return to
this topic.
--
Maxim Dounin
http://mdounin.ru/
_______________________________________________
nginx-devel mailing list
nginx-devel@nginx.org
http://mailman.nginx.org/mailman/listinfo/nginx-devel