[Libguestfs] Benchmarks of asynch nbdublk (was: Re: [PATCH libnbd] ublk: Add new nbdublk program)

[Richard W.M. Jones]

On Sun, Aug 28, 2022 at 04:39:07PM +0100, Richard W.M. Jones wrote:
> Performance was within 25% of disk access over virtio-scsi, which I
> think is pretty good.

BTW we had a GSoC student last summer who was supposed to work on
adding io_uring support to libnbd.  Unfortunately they didn't complete
the assignment (or start it, to be honest).  We would still like to
add it to libnbd in future.

However libnbd implements a network protocol and while io_uring would
certainly overcome the syscall KPTI overhead, there's still the basic
problem that unlike block I/O, network I/O has dependencies between
requests - you cannot execute them out of order - so although it'll
have some benefit I don't think it'll make a huge difference.  (Be
interesting to find out whether or not I'm right about this.)

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
nbdkit - Flexible, fast NBD server with plugins
https://gitlab.com/nbdkit/nbdkit