VMX Hardware

VMX Hardware

[Goetz Bock]

Sorry for the cross post.

My server finaly quitted to work :-(, but I had planed to upgrade at the
end of the year anyway.

As the VMX hardware is supposed to be out "soon" I actually wanted to
wait for it to hit the shelvs first.

Looks like I can not do this. so I've a question for the informed.

What hardware should I get now, and still be able to upgrade to VMX when
it's available? If possible only by getting a new CPU.

I was planing for either a Pentium D and Athlon X2 (prefered), the low
end server market.

My server used to run 8 DomUs (it was a Pentium 4, 2.4GHz /w HT) but the 
new one should be able to handle about 16.
-- 
Goetz Bock       (c) 2005 as     blacknet.de - Munich - Germany   /"\
IT Consultant  Creative Commons  secure mobile Linux everNETting  \ /
                                                                   X
 ASCII Ribbon Campaign against HTML email & microsoft attachments / \

Re: VMX Hardware

[Keir Fraser]

On 14 Aug 2005, at 11:44, Goetz Bock wrote:

> As the VMX hardware is supposed to be out "soon" I actually wanted to
> wait for it to hit the shelvs first.
>
> Looks like I can not do this. so I've a question for the informed.
>
> What hardware should I get now, and still be able to upgrade to VMX 
> when
> it's available? If possible only by getting a new CPU.

Buying a mainboard with LGA775 socket is probably a fairly safe bet, 
but there's no guarantee that future processor generations will run on 
it even if the socket doesn't change. Compatibility might depend on the 
manufacturer creating appropriate BIOS patches, for example.

  -- Keir

Re: Re: [Xen-devel] VMX Hardware

[Mogens Valentin]

Keir Fraser wrote:
> 
> On 14 Aug 2005, at 11:44, Goetz Bock wrote:
> 
>> As the VMX hardware is supposed to be out "soon" I actually wanted to
>> wait for it to hit the shelvs first.
>>
>> Looks like I can not do this. so I've a question for the informed.
>>
>> What hardware should I get now, and still be able to upgrade to VMX when
>> it's available? If possible only by getting a new CPU.
> 
> 
> Buying a mainboard with LGA775 socket is probably a fairly safe bet, but 
> there's no guarantee that future processor generations will run on it 
> even if the socket doesn't change. Compatibility might depend on the 
> manufacturer creating appropriate BIOS patches, for example.

Looking at Intels latest 'upgrade' path, which shows we'll often have to 
buy new HW for new cpu's, I'd go for AMD.
Seems they're into the game of longtime support, so my bet is that 
you'll be able to outfit quite a lot of the latest mobos with future 
AMD64 cpu's.
This has of late been the case with i.e. their X2 line. Most resent 
mobos can handle those with a BIOS upgrade.
I'd be looking at Iwill, Tyan, Gigabye.

I assume by VMX, you're talking about cpu/hardware with HW-support for 
virtualisation, i.e. Intels Vanderbilt Technology (VT) and AMD' 
Pacifica. If so, it's clear to me that AMD will be leading, because of 
it's onchip memory controller, and hence reduced needs for software 
support to do the page switching, for which Intel needs quite a lot of 
software to emulate.

However, I'm not sufficiently familiar with Xen on this topic (just 
getting started). What I've read is that Xen won't benefit from or even 
use those HW-enabled mechanisms.
Someone pls. correct me if I'm wrong here - I'd very much like to be 
wrong. Would be a pity not to take advantage of such mechanisms, IMHO.

-- 
Kind regards,
Mogens Valentin

Re: [Xen-users] Re: VMX Hardware

[Keir Fraser]

On 14 Aug 2005, at 13:23, Mogens Valentin wrote:

> I assume by VMX, you're talking about cpu/hardware with HW-support for 
> virtualisation, i.e. Intels Vanderbilt Technology (VT) and AMD' 
> Pacifica. If so, it's clear to me that AMD will be leading, because of 
> it's onchip memory controller, and hence reduced needs for software 
> support to do the page switching, for which Intel needs quite a lot of 
> software to emulate.

Not sure what you mean. Both technologies require the hypervisor to 
maintain shadow page tables. SVM (Pacifica) does have the advantage of 
a tagged TLB, so avoiding TLB flushes when switching between guest mode 
and hypervisor mode, and between guests, but we have no measurements so 
estimates of relative performance are pure conjecture.

> However, I'm not sufficiently familiar with Xen on this topic (just 
> getting started). What I've read is that Xen won't benefit from or 
> even use those HW-enabled mechanisms.
> Someone pls. correct me if I'm wrong here - I'd very much like to be 
> wrong. Would be a pity not to take advantage of such mechanisms, IMHO.

Xen will use VT and SVM to run unmodified operating systems. This will 
be useful for virtualising legacy system installations, and for 
supporting OSes where the source is not freely available.

Xen already has VT support, and I expect that SVM will also be 
supported by the end of the year. Hardware won't be generally available 
before then anyway.

  -- Keir

Re: Re: [Xen-devel] VMX Hardware

[Anthony Liguori]

Keir Fraser wrote:

>
> On 14 Aug 2005, at 13:23, Mogens Valentin wrote:
>
>> I assume by VMX, you're talking about cpu/hardware with HW-support 
>> for virtualisation, i.e. Intels Vanderbilt Technology (VT) and AMD' 
>> Pacifica. If so, it's clear to me that AMD will be leading, because 
>> of it's onchip memory controller, and hence reduced needs for 
>> software support to do the page switching, for which Intel needs 
>> quite a lot of software to emulate.
>
>
> Not sure what you mean. Both technologies require the hypervisor to 
> maintain shadow page tables.

The Pacifica spec does mention optional hardware support for "Nested 
Page Tables" which is essentially hardware shadow page tables.

Should be interesting to see the performance difference between 
software/hardware shadow page tables.

It's worth noting that AMD has just released a simulator for (SimNow) 
that has virtualization support.  Perhaps one of the AMD guys on the 
list can confirm whether SimNow has enough Pacifica support for Xen 
development?

Regards,

Anthony Liguori

>
>  -- Keir
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
>

Re: [Xen-users] Re: VMX Hardware

[Keir Fraser]

On 14 Aug 2005, at 19:09, Anthony Liguori wrote:

> The Pacifica spec does mention optional hardware support for "Nested 
> Page Tables" which is essentially hardware shadow page tables.
>
> Should be interesting to see the performance difference between 
> software/hardware shadow page tables.

I have doubts about the performance of hardware page tables, unless 
there's a clever TLB design. A pagetable walk which is four memory 
accesses on normal x86/64 becomes twenty accesses in nested mode!

It may be that shadow pagetables, by providing a bigger 'cache' for 
guest->host mappings, can actually perform better on many workloads.

I guess we'll see -- if silicon is produced that supports that mode, 
someone is bound to add support to Xen.

  -- Keir

Re: VMX Hardware

[David]

Hi,

 > Looking at Intels latest 'upgrade' path, which shows we'll often have 
to buy new HW for new cpu's, I'd go for AMD.
 > Seems they're into the game of longtime support, so my bet is that 
you'll be able to outfit quite a lot of the latest mobos with future 
AMD64 cpu's.
 > This has of late been the case with i.e. their X2 line. Most resent 
mobos can handle those with a BIOS upgrade.
 > I'd be looking at Iwill, Tyan, Gigabye.
 >

This is conjecture, nor am I well versed in this technology, so take 
this with a grain of salt.

I would assuume that both Intel and AMD platforms will need a new 
mainboard for the switch to hardware VM tech...

Why?

Intel: If only beacsue if the external memory controller, seemingly they 
Intel will need a new chipset.  I also don't see Intel passing on an 
oportuniity to sell more chipsets.

AMD:  With AMD planning to rollout support for DDR-2 and sockets M2/F, 
it is unlikely that hardware VM tech will be back-ported to the current 
generation A64 platform.

-- David