RE: [Xen-users] VT or paravirtualization?

["Petersson, Mats" <Mats.Petersson@xxxxxxx>]

> -----Original Message-----
> From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx 
> [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of 
> Czakó Krisztián
> Sent: 29 January 2006 21:03
> To: xen-users@xxxxxxxxxxxxxxxxxxx
> Subject: [Xen-users] VT or paravirtualization?
> 
> Hello,
> 
> What is the better solution now to run linux domUs under Xen3?
> Using a VT hardware, sould I install a xen-patched kernel as 
> I would where no VT extension, or use an unmodified linux 
> guest system?

At this time, I'd say that Para-virtualization is better performing. This is 
based on the fact that we're currently on the first release of Hardware 
virtualization, and there's several things that are less than optimal. Device 
model optimisation for specific devices would be one clear benefit, but there's 
other possibilities too. 

Obviously, if you want to run unmodified OS's that doesn't have freely 
available source, or perhaps run an application that REQUIRES a particular 
distribution that can't have a modified kernel, then full virtualization is 
pretty much the only option.

> 
> Is there any performance difference between a Pentium D 3 Ghz with vt
> (9xx) and without it (8xx)? 

I can't speak for Intel, but from AMD's perspective [1] I would expect there to 
be NO difference for your average application software - kernel mode 
instructions such as control register access and such things MAY be a tiny bit 
slower (but these are generally VERY EXPENSIVE instructions ANYWAYS, for 
example a move to CR3 will invalidate the TLB's on the processor, which means 
that the very next memory read will require reading of all levels of the 
page-table [ignoring the fact that there's optimisations in the processor that 
"ignore" moves to CR3 with the same value]).

I'd doubt that any REAL application benchmark would be able to show any 
difference that is statistically verifiable. 

[1] I'm not a processor architect, so this is MY _OPINION_, rather than a 
statement of facts from AMD as a corporation.

--
Mats
> 
> Regards,
> Slapic
> 
> 
> 
> _______________________________________________
> Xen-users mailing list
> Xen-users@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-users
> 
> 


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