[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: [Libguestfs] [RFC PATCH] protocol: Add NBD_CMD_FLAG_FAST_ZERO

On Fri, Mar 22, 2019 at 12:17:59PM -0500, Eric Blake wrote:
> On 3/22/19 11:42 AM, Eric Blake wrote:
> > 
> > Hence, it is desirable to have a way for clients to specify that a
> > particular write zero request is being attempted for a fast wipe, and
> > get an immediate failure if the zero request would otherwise take the
> > same time as a write.  Conversely, if the client is not performing a
> > pre-initialization pass, it is still more efficient in terms of
> > networking traffic to send NBD_CMD_WRITE_ZERO requests where the
> > server implements the fallback to the slower write, than it is for the
> > client to have to perform the fallback to send NBD_CMD_WRITE with a
> > zeroed buffer.
> > 
> > Add a protocol flag and corresponding transmission advertisement flag
> > to make it easier for clients to inform the server of their intent. If
> Note that this is independent of proposals made on the NBD list in the
> past [1] of having a way for the server to advertise that a particular
> export starts in an all-zeroes state (faster than a series of 32-bit
> NBD_CMD_BLOCK_STATUS would be able to do), although I may _also_ try to
> revive proposed documentation and a reference implementation of that
> optimization as well (as qemu-img convert can completely skip the
> zeroing, whether the bulk wipe or per-hole writing, when it knows the
> destination is already zero).

It has to be said that this would be a lot easier to implement, and
for our purposes (optimizing qemu-img convert) it does everything we

However the original proposal you put here seems reasonable.  I have
only one comment about it: Should the new error (ENOTSUP) be submitted
as a separate patch to the spec?

> [1] https://lists.debian.org/nbd/2016/12/msg00015.html and following
> (doc: Propose NBD_FLAG_INIT_ZEROES extension)
> > 
> > I will not push this without both:
> > - a positive review (for example, we may decide that burning another
> > NBD_FLAG_* is undesirable, and that we should instead have some sort
> > of NBD_OPT_ handshake for determining when the server supports

>From an implementation point of view I prefer simple flags over having
to implement a brand new option.

We can always work out how to extend the flags field if we run out of
flags.  For example, by implementing NBD_OPT_INFO2 with a much bigger
flags field.

> > - a reference client and server implementation (probably both via qemu,
> > since it was qemu that raised the problem in the first place)
> The last time we mentioned the possibility of advertising an initial
> zero state, we debated whether burning one of our 16 NBD_FLAG_*
> transmission bits for that purpose was wise [2], but discussion stalled
> after I never developed a counterproposal with NBD_OPT_* handshaking and
> never produced a reference implementation.
> [2] https://lists.debian.org/nbd/2016/12/msg00048.html
> Also, keep in mind that knowing that something started as all zeroes
> (which only affects startup; once you do any write, that early status
> bit no longer means anything to current operation, so less important to
> hand to the kernel during transmission phase, especially if the kernel
> can ever learn to utilize NBD_CMD_BLOCK_STATUS) is indeed different from
> knowing if probing for fast zeroing is supported (where it is easy to
> conceive of the kernel needing to know if it can send
> NBD_CMD_FLAG_FAST_ZERO).  So we may still want to use NBD_OPT_* to get
> the initial zero extension, but NBD_FLAG to advertise the fast zero
> extension.
> On the other hand, it's also worth thinking about which extensions are
> easy for servers to implement - NBD_FLAG_INIT_ZEROES and
> NBD_FLAG_SEND_FAST_ZERO are orthogonal enough that I could see a full
> 2x2 mix of servers (unsupported, either one of the two supported, or
> both supported), and where clients may make optimization choices based
> on any of those four combinations.
> [and if we're keeping score, other extension proposals that I want
> revisit, in no particular order, include:
> - 64-bit operations
> - more precision on TRIM/WRITE_ZERO alignment constraints
> ]


Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
Fedora Windows cross-compiler. Compile Windows programs, test, and
build Windows installers. Over 100 libraries supported.

Reply to: