Re: [dpdk-dev] [PATCH] vhost: improve dirty pages logging performance

[Tiwei Bie <tiwei.bie@intel.com>]

On Tue, May 15, 2018 at 03:50:54PM +0200, Maxime Coquelin wrote:
> Hi Tiwei,
> 
> I just see I missed to reply to your comment on my commit message:
> 
> On 05/03/2018 01:56 PM, Tiwei Bie wrote:
> > On Mon, Apr 30, 2018 at 05:59:54PM +0200, Maxime Coquelin wrote:
> > > This patch caches all dirty pages logging until the used ring index
> > > is updated. These dirty pages won't be accessed by the guest as
> > > long as the host doesn't give them back to it by updating the
> > > index.
> > Below sentence in above commit message isn't the reason why
> > we can cache the dirty page logging. Right?
> > 
> > """
> > These dirty pages won't be accessed by the guest as
> > long as the host doesn't give them back to it by updating the
> > index.
> 
> That's my understanding.
> As long as the used index is not updated, the guest will not process
> the descs.
> If the migration converges between the time the descs are written,
> and the time the used index is updated on source side. Then the guest
> running on destination will not see the descriptors as used but as
> available, and so will be overwritten by the vhost backend on
> destination.

If my understanding is correct, theoretically the vhost
backend can cache all the dirty page loggings before it
responds to the GET_VRING_BASE messages. Below are the
steps how QEMU live migration works (w/o postcopy):

1. Syncing dirty pages between src and dst;
2. The dirty page sync converges;
3. The src QEMU sends GET_VRING_BASE to vhost backend;
4. Vhost backend still has a chance to log some dirty
   pages before responding the GET_VRING_BASE messages;
5. The src QEMU receives GET_VRING_BASE response (which
   means the device has stopped);
6. QEMU sync the remaining dirty pages;
7. QEMU on the dst starts running.

(The steps 3~6 are the downtime which we want to minimize)

So I think the words in commit log isn't really related
to why we can cache the dirty page loggings.

Best regards,
Tiwei Bie

> 
> Regards,
> Maxime