Re: [Qemu-devel] [Nbd] write_zeroes/trim on the whole disk

[Wouter Verhelst]

On Sat, Sep 24, 2016 at 11:31:25AM +0100, Alex Bligh wrote:
> 
> > On 23 Sep 2016, at 22:21, Wouter Verhelst <address@hidden> wrote:
> > 
> > On Fri, Sep 23, 2016 at 02:00:06PM -0500, Eric Blake wrote:
> >> My preference would be a new flag to the existing commands, with
> >> explicit documentation that 0 offset and 0 length must be used with that
> >> flag, when requesting a full-device wipe.
> > 
> > Alternatively, what about a flag that says "if you use this flag, the
> > size should be left-shifted by X bits before processing"? That allows
> > you to do TRIM or WRITE_ZEROES on much larger chunks, without being
> > limited to "whole disk" commands. We should probably make it an illegal
> > flag for any command that actually sends data over the wire, though.
> 
> I'd prefer an approach like this. Perhaps X could be negotiated
> with the block size extension (or simply be defined as the
> 'preferred block size'.
> 
> This could then be defined to work with all commands.

The downside of making the shift size too small is that it wouldn't help
as much with the use case that was presented here (that is, blanking a
whole disk with one command). I think if the shift size is to be
negotiated, it should be something that the client prefers, rather than
something the server prefers.

I really don't think allowing a shift for things that requires data
transfer makes much sense. Yes, a shift for things like "clear all data"
can be sensible, because (depending on backend) this can be a fairly
simple operation. However, "write more than 2G of data" or "read more
than 2G of data" isn't a simple operation and takes a lot of data to be
transferred, so allowing shift for those operations doesn't make sense,
at all; and suggesting that we allow it could actually make people
believe it's a good idea to do so when it really really (really) isn't.

-- 
< ron> I mean, the main *practical* problem with C++, is there's like a dozen
       people in the world who think they really understand all of its rules,
       and pretty much all of them are just lying to themselves too.
 -- #debian-devel, OFTC, 2016-02-12