On Thu, Sep 24, 2020 at 12:27:14PM +0200, SeongJae Park wrote:

On Thu, 24 Sep 2020 12:13:44 +0200 "Roger Pau Monné" [off-list ref] wrote:

quoted

On Wed, Sep 23, 2020 at 04:09:30PM -0400, Konrad Rzeszutek Wilk wrote:

quoted

On Tue, Sep 22, 2020 at 09:01:25AM +0200, SeongJae Park wrote:

quoted

From: SeongJae Park <redacted>

Persistent grants feature provides high scalability.  On some small
systems, however, it could incur data copy overhead[1] and thus it is
required to be disabled.  But, there is no option to disable it.  For
the reason, this commit adds a module parameter for disabling of the
feature.

Would it be better suited to have it per guest?

I think having a per-backend policy that could be specified at the
toolstack level would be nice, but I see that as a further
improvement.

Agreed.

quoted

Having a global backend domain policy of whether persistent grants are
enabled or not seems desirable, and if someone wants even more fine
grained control this change is AFAICT not incompatible with a
per-backend option anyway.

I think we could extend this design by receiving list of exceptional domains.
For example, if 'feature_persistent' is True and exceptions list has '123,
456', domains of domid 123 and 456 will not use persistent grants, and vice
versa.

I think that would be quite fragile IMO, I wouldn't recommend relying
on domain IDs.

What I would do instead is add a new attribute to
xl-disk-configuration [0] that allows setting the persistent grants
usage on a per-disk basis, and that should be passed to blkback in a
xenstore node.

I could implement this, but... to be honest, I don't really understand the
needs of the fine-grained control.  AFAIU, the problem is 'scalability' vs
'data copy overhead'.  So, only small systems would want to turn persistent
grants off.  In such a small system, why would we need fine-grained control?
I'm worrying if I would implement and maintain a feature without real use case.

For the reason, I'd like to suggest to keep this as is for now and expand it
with the 'exceptions list' idea or something better, if a real use case comes
out later.

I agree. I'm happy to take patches to implement more fine grained
control, but that shouldn't prevent us from having a global policy if
that's useful to users.

Roger.

[0] https://xenbits.xen.org/docs/unstable/man/xl-disk-configuration.5.html