Re: [Qemu-devel] [PATCH v1 00/17] dataplane: optimization and multi virtqueue support

[Kevin Wolf]

Am 17.08.2014 um 07:29 hat Paolo Bonzini geschrieben:
> Il 15/08/2014 22:15, Paolo Bonzini ha scritto:
> >> >                 | Random throughput | Sequential throughput
> >> > ----------------+-------------------+-----------------------
> >> > master          | 442 MB/s          | 730 MB/s
> >> > base            | 453 MB/s          | 757 MB/s
> >> > bypass (Ming)   | 461 MB/s          | 734 MB/s
> >> > coroutine       | 468 MB/s          | 716 MB/s
> >> > bypass (Paolo)  | 476 MB/s          | 682 MB/s
> > This is pretty large, but it really smells like either a setup problem
> > or a kernel bug...
> 
> Thinking more about the I/O scheduler, it could simply be that faster
> I/O = less coalescing = more bios actually reaching the driver = less speed.
> 
> It should be possible to find if this is true using blktrace.
> 
> (The reason why sequential I/O is faster is coalescing in the I/O
> scheduler).

Yes, sorry, should have posted an update on Friday. This was cfq in the
guest (apparently, the host doesn't use any scheduler with loop devices,
which makes some sense), with noop (or deadline) the numbers look much
better: Sequential throughput is on a slightly higher level and
increases with the optimisations, and random is much closer to it.

Kevin