Il 15/03/2012 01:42, Richard Laager ha scritto:
>> > It is worse in that we do not want the hardware parameters exposed to the
>> > guest to change behind the scenes, except if you change the machine type
>> > or if you use the default unversioned type.
> You're saying that discard_granularity and discard_zeros_data need to be
> properties of the machine type? If you start with that as a requirement,
> I can see why you want to always report discard_granularity=512 &
> discard_zeros_data=1. But that design has many downsides. I'm not
> convinced that discard_granularity and discard_zeros_data need to be
> properties of the machine type. Why do you feel that's necessary? What's
> the harm in those properties changing across QEMU startups (i.e. guest
> boots)?

Changing across guest boots is a minor problem, but changing across
migration must be avoided at all costs.

BTW, after this discussion I think we can instead report
discard_granularity = 512 and discard_zeroes_data=0 and get most of the
benefit, at least on file-backed storage.

Paolo

Reply via email to