Re: vnuma toolstack extension request

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Wed, Sep 24, 2014 at 06:09:24PM +0200, Dario Faggioli wrote:
> On mer, 2014-09-24 at 11:24 -0400, Elena Ufimtseva wrote:
> 
> > 
> > On Wed, Sep 24, 2014 at 11:15 AM, Konrad Rzeszutek Wilk
> > <konrad.wilk@oracle.com> wrote:
> 
> >         Hey Elena,
> > 
> > 
> > Hi Konrad
> >
> Hi Elena and Konrad,
>  
> >         Before you invest time in this, a couple of questions:
> >         
> >          - Dariof was interested in merging this with his auto NUMA
> >         placement code.
> >            That piece of code does not look simple and I think he
> >         volunteered
> >            to do it. 
> >
> Not a piece of cake, definitely, mostly because of the many possible
> ways the two things (vNUMA topology and automatic placement) can
> interact and influence each others.
> 
> >         However he has also been busy reviewing Xen RT patches so
> >            I figured he hasn't yet done that.
> >  
> Exactly. I'm actually mostly done with it, but it needs a bit of
> refining of a couple of rough edges, and for sure some testing, and,
> once posted, integration with the other toolstack changes Elena has done
> while addressing review comments, and then of course re-reviewing.
> 
> > Yes, Dario mentioned he will be working on this.
> >         
> And I was, but as Konrad guessed, reviewing and testing the RT
> scheduling stuff ate some time.
> 
> Also, consider that I'm about to disappear for at least two weeks,
> starting (most likely) from next Monday, for (good :-) ) family reasons.
> That means I probably won't be able to review/help with any future
> version of the series.
> 
> Of course I'm no maintainer, I know, so that is not a showstopper
> per-se. I'm just saying I think I could have been of help, while I
> probably won't. :-)
> 
> >          - Since it would interact with auto NUMA placement and expose
> >         new
> >            logic in libxl we run in the problem of the 'stable' API
> >         that we
> >            MUST preserve. That is a worry - and will require
> >         scrutinity.
> >         
> Indeed.
> 
> >          - What is the impact if the toolstack patches go in Xen 4.6?
> >
> FWIW, my personal opinion is that we better actually defer this to 4.6.
>  
> > Maybe maintainers will have more info about this. I am only aware of
> > that
> > vNUMA will not be supported by toolstack in 4.5, so some related work
> > will not be merged 
> > as vNUMA aware ballooning and HVM support (not sure if the last one is
> > happening though).
> >
> Yes, true, this is a rather basic building block of more general vNUMA
> support. However, this also mean that, even if we merge it, we would
> only have part of the picture in place for 4.5 anyway. I.e., we can work
> on 4.6 to have all the vNUMA bits and pieces, and make a really big fuss
> about that when releasing! :-)
> 
> This does not mean I would not have liked to have it there (given all
> the effort that Elena, and recently Wei, put into it), but I guess this
> is how software development work, isn't it? :-/
> 
> So, in summary, my take on this is: a lot of progress has been made,
> which was not obvious, at least at some point. It's not yet ready, but
> that only mean 4.6 will be a great release wrt vNUMA!! :-D


Yes! And it will give Elena some time to relax instead of hurridly have
to code and test all of this :-)

Elena, I am sorry - but based on the feedback I believe the best
course is to defer this to Xen 4.6.

> 
> Regards,
> Dario
> 
> -- 
> <<This happens because I choose it to happen!>> (Raistlin Majere)
> -----------------------------------------------------------------
> Dario Faggioli, Ph.D, http://about.me/dario.faggioli
> Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
>