On 11/08/2010 08:29 PM, Andrew Cathrow wrote:
----- Original Message -----
> From: "Dennis Jacobfeuerborn"<dennisml(a)conversis.de>
> To: "Andrew Cathrow"<acathrow(a)redhat.com>
> Cc: "Bill Davidsen"<davidsen(a)tmr.com>, xen(a)lists.fedoraproject.org,
virt(a)lists.fedoraproject.org, "M A Young"
> <m.a.young(a)durham.ac.uk>
> Sent: Monday, November 8, 2010 1:59:54 PM
> Subject: Re: [fedora-virt] [Fedora-xen] Dom0 xen support in Fedora 15?
> On 11/08/2010 06:02 PM, Andrew Cathrow wrote:
>>
>>
>> ----- Original Message -----
>>> From: "Bill Davidsen"<davidsen(a)tmr.com>
>>> To: dlaor(a)redhat.com
>>> Cc: xen(a)lists.fedoraproject.org, virt(a)lists.fedoraproject.org, "M A
>>> Young"<m.a.young(a)durham.ac.uk>
>>> Sent: Monday, November 8, 2010 11:52:08 AM
>>> Subject: Re: [fedora-virt] [Fedora-xen] Dom0 xen support in Fedora
>>> 15?
>>> Dor Laor wrote:
>>>> On 11/08/2010 04:55 AM, M A Young wrote:
>>>>
>>>>> I am trying to work out whether it is practical to propose Dom0
>>>>> xen
>>>>> support as a feature for Fedora 15.
>>>>>
>>>>> The kernel situation is that Domain 0 has been accepted upstream
>>>>> for
>>>>> 2.6.37. Assuming a 3 month kernel release cycle, F15 will most
>>>>> likely ship
>>>>> with a 2.6.37.x kernel, with 2.6.38 coming out either after the
>>>>> F15
>>>>> release or just before but too late to be included. If the plan
>>>>> to
>>>>> get key
>>>>> xen drivers into 2.6.38 succeeds, then F15 may be become usable
>>>>> as
>>>>> a
>>>>> Domain 0 system at some point during its lifetime as the kernel
>>>>> package in
>>>>> a Fedora version typically has one major update.
>>>>>
>>>>> If the kernel team accept backported patches then it might just
>>>>> be
>>>>> possible to ship F15 with usable Domain 0 support but the
>>>>> timescale
>>>>> for
>>>>> that would be very tight.
>>>>>
>>>>> The other thing we would need to consider is what needs to be
>>>>> done
>>>>> to make
>>>>> xen friendly enough to be usable by an ordinary user. The page
>>>>>
https://fedoraproject.org/wiki/Features/XenPvopsDom0 contains
>>>>> plans
>>>>> from
>>>>> when dom0 xen support was expected to make a quick return to
>>>>> Fedora, but
>>>>> they are a couple of years old now so probably need updating.
>>>>>
>>>>> I think as a minimum we would need a way to add a dom0 enabled
>>>>> grub
>>>>> entry
>>>>> for a kernel, rather than requiring the user to hand edit the
>>>>> grub
>>>>> file.
>>>>> We should also make sure that xen works with the other Fedora
>>>>> virtualisation tools.
>>>>>
>>>>> What do others think about this? For example is it achievable as
>>>>> a
>>>>> feature, is it too early and better to wait for F16, and what
>>>>> else
>>>>> should
>>>>> we aim to do to make xen usable in Fedora?
>>>>>
>>>> Have you consider kvm? it's upstream since 2.6.20 and now its more
>>>> ready
>>>> than ever.
>>>>
>>>
>>> There are some good tutorials which should explain the difference
>>> between xen and kvm, particularly the performance and hardware
>>> requirements of each.
>>
>> re: hardware requirements, KVM's requirement for VT-X/AMD-V
>> extensions certainly used to be a concern 2-3 years ago but today
>> even laptops come with this support.
>> And regarding performance they days of Xen outperforming KVM have
>> long-since passed.
>
> Citations needed. I'm not saying what you claim isn't true but without
> data
> this opinion doesn't carry much weight.
>
Citations are really needed on both sides of the debate, 2 or 3 year old metrics no
longer apply.
Which is why I hoped I could nudge you into providing some data if you had
any. :)
There seems to be very little recent and hard data on the subject out there.
Vendors published benchmarks are typically questionable, they focus
on their products strengths and their competitors weakness.
The only hope for a fair comparison is a vendor neutral set of benchmarks such as
SPECvirt
http://www.spec.org/virt_sc2010/
But obviously this isn't a simple test to run.
AFAIK there are some issues with that benchmark not being free.
What I'm basically looking for is a simple comparison of a few key metrics
between common setups. For example a simple bonnie++ run on XenPV and
KVM+virtio_blk could already give people at least some baseline numbers.
Regards,
Dennis