[Xen-ia64-devel] RE: Implementing both (was: Xen/ia64 - global or per VP VHPT)

["Magenheimer, Dan (HP Labs Fort Collins)" <dan.magenheimer@xxxxxx>]

I _was_ proposing that both mechanisms be present at run time,
but not necessarily permanently -- just to allow easier
testing/comparison during development. By the time we have
real users doing lots of domains, I would hope we would
have enough data to make a decision.

Compile-time option would be OK too, but I suspect it
will lead to a debate about what the distribution-default
should be :-) or it will keep changing back and forth
depending on who did the last checkin :-)

Either way is OK with me.

> -----Original Message-----
> From: Munoz, Alberto J [mailto:alberto.j.munoz@xxxxxxxxx] 
> Sent: Monday, May 02, 2005 7:51 PM
> To: Magenheimer, Dan (HP Labs Fort Collins); ipf-xen; 
> xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
> Subject: RE: Implementing both (was: Xen/ia64 - global or per VP VHPT)
> 
> Hi Dan,
> 
> The first thing I want to ask is whether or not you are 
> proposing that both
> mechanisms be present at run time. I definitely don't think 
> that is the
> right thing to do, as then we will have the worst of both 
> worlds, a global
> VHPT that needs to be dimensioned to the full size of memory 
> and the per VM
> VHPTs that would be requiring additional memory.
> 
> I have no objection to both solutions being implemented in 
> the sources and
> having the ability to pick one or the other (through #ifdef). 
> This should
> make it easy to compare which one does better (in most cases) 
> and would make
> it very easy to drop the one that does not.
> 
> Bert
> 
> Magenheimer, Dan (HP Labs Fort Collins) 
> <mailto:dan.magenheimer@xxxxxx>
> wrote on Monday, May 02, 2005 4:34 PM:
> 
> > Thanks, Bert, for the good summary of issues.
> > 
> >> If everyone agrees that doing both
> >> implementations in the same source base is feasible and does
> >> not adversely
> >> affect other stuff, then I have no objection to what you propose.
> > 
> > Let's start the feasibility discussion on a new base thread...
> > 
> > I am probably oversimplifying it, but I'm not sure it will be
> > that hard to support both approaches.  Domain0's VHPT is set
> > up at boot time.  When new domains are launched they should
> > specify whether they want to "share" Domain0's VHPT (global
> > VHPT) or have their own VHPT allocated (per-domain VHPT).
> > For now, the default for paravirt domains can be shared and
> > the default (or, if necessary, the ONLY choice) for VT domains
> > can be per-domain.
> > 
> > The choice (and the location of the physical IVA) becomes
> > part of the per-domain state.  At domain switch time, cr.iva
> > is changed as necessary.
> > 
> > Now everybody can commence shooting holes in my 
> oversimplification :-)
> > 
> > Dan
> 
> Bert
> 

_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel