[ovs-dev] [PATCH v4] datapath-windows: Enable checksum offloads in STT

[Ben Pfaff]

On Tue, Sep 29, 2015 at 01:05:33PM -0700, Jesse Gross wrote:
> On Tuesday, September 29, 2015, Ben Pfaff <blp at nicira.com
> <javascript:_e(%7B%7D,'cvml','blp at nicira.com');>> wrote:
> 
> > On Wed, Sep 23, 2015 at 08:49:49AM -0700, Jesse Gross wrote:
> > > On Wed, Sep 23, 2015 at 8:38 AM, Nithin Raju <nithin at vmware.com> wrote:
> > > >> On Sep 23, 2015, at 8:33 AM, Jesse Gross <jesse at nicira.com> wrote:
> > > >>
> > > >> On Tue, Sep 22, 2015 at 11:04 PM, Nithin Raju <nithin at vmware.com>
> > wrote:
> > > >>> hi Jesse,
> > > >>> We are getting the Hyper-V solution to a state with the following
> > goals:
> > > >>> - Work “out of the box” ie. no need to make special settings such as
> > disabling checksum offload, TSO, etc.
> > > >>> - Reasonably stable
> > > >>>
> > > >>> Most of the patches we have checked in so far into 2.4 are geared
> > towards these two goals. Once all of the required changes go in, and we are
> > reasonably confident about the stability, we can hopefully make an
> > announcement about Hyper-V support.
> > > >>
> > > >> This is my concern - there should not be any announcements based on
> > > >> stable branches because there should be no development on stable
> > > >> branches. The only thing that should go in is targeted bug fixes to
> > > >> address issues that came up after the release.
> > > >>
> > > >> All of the goals that you listed are good things and make sense - on
> > > >> the master branch. However, I don't see a need to bring these back to
> > > >> 2.4. My guess is that there is no more churn in the common code on
> > > >> master than with the Windows patches here.
> > > >>
> > > >> So please just target 2.5 as the release to make an announcement about
> > > >> Hyper-V support. I promise that this release cycle won't be as long as
> > > >> 2.4.
> > > >
> > > > Jesse,
> > > > We were hoping for a dot release off of 2.4. Like a 2.4.1 or 2.4.2 to
> > announce support. Would that not be the right release vehicle?
> > >
> > > No, there should be no new features in point releases. 2.5 is the next
> > > release where it would makes sense to do this. My guess is that will
> > > be some time around the end of the year.
> >
> > The patches I've seen from the Hyper-V developers so far are just in
> > Hyper-V specific code, that can't really affect the stability of the
> > rest of the platform.  I have questions about the value of doing this on
> > 2.4, given that 2.5 will branch in a reasonable amount of time, but
> > since they're eager to do it I'm not sure that it's worth discouraging
> > them ;-)
> >
> 
> I think it does cause problems for people who want to upgrade OVS in
> production deployments as often it is not desirable to just take a large
> set of changes wholesale for a bugfix release. If there are many unrelated
> commits for feature development it makes it harder to identify what is
> actually going on. I'm also not sure that it is true that there are no
> changes to common code.

Nithin, Alin, et al., why do you guys want this on branch-2.4 so badly?
Everything that Jesse is saying is right, and despite my initial
instincts I'm inclined to agree with him in the end.