Re: [Qemu-devel] [PATCH 0/3] block: introduce submit I/O at batch

[Paolo Bonzini]

Il 30/06/2014 11:49, Ming Lei ha scritto:
> Hi,
>
> The commit 580b6b2aa2(dataplane: use the QEMU block layer for I/O)
> introduces ~40% throughput regression on virtio-blk dataplane, and
> one of causes is that submitting I/O at batch is removed.
>
> This patchset trys to introduce this mechanism on block, at least,
> linux-aio can benefit from that.
>
> With these patches, it is observed that thoughout on virtio-blk
> dataplane can be improved a lot, see data in commit log of patch
> 3/3.
>
> It should be possible to apply the batch mechanism to other devices
> (such as virtio-scsi) too.

The basic idea of the code is great, however I do not think it is 
necessary to add the logic in AioContext.  You are right that io_submit 
supports multiple I/O, but right now we have one io_context_t per 
BlockDriverState so it is somewhat premature.  Note that AioContext in 
QEMU means something else than io_context_t.

I don't think it's necessary to walk backing_hd too (especially because 
we're close to release and this cannot be a regression---dataplane in 
2.0 didn't support backing files at all!).  We need to keep the patches 
as simple as possible.

You can just add bdrv_plug/bdrv_unplug callbacks to block/linux-aio.c 
and, in block.c, something like

    BlockDriver *drv = bs->drv;
    if (drv && drv->bdrv_plug) {
        drv->bdrv_plug(bdrv);
    } else if (bs->file) {
        bdrv_plug(bs->file);
    }

and place all the queuing logic in linux-aio.c.  It should be obvious to 
the reviewer that the patch only affects dataplane.

(Also, note that QEMU doesn't use __ as the prefix for internal function).

Thanks for the patch!

Paolo