state of xenified SUSE kernel

state of xenified SUSE kernel

[Andreas Kinzler]

Hello Jan,

perhaps you can shed some light on the state of the xenfied SUSE kernels 
(http://kernel.opensuse.org/cgit/kernel-source).

We use xenified kernels based on kernel 3.4 for years and benchmarks 
showed that they are faster than the pvops (vanilla) kernels.
But what is the current state in terms of performance and features?

Related question is if they have a future (at SUSE) at all? The repo 
does not have Xen patches for SLE12-SP2. Still coming? Or dropped forever?

Regards Andreas

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: state of xenified SUSE kernel

[Jan Beulich]

>>> On 17.05.16 at 17:10, <ml-ak@posteo.de> wrote:
> We use xenified kernels based on kernel 3.4 for years and benchmarks 
> showed that they are faster than the pvops (vanilla) kernels.
> But what is the current state in terms of performance and features?

I'm not sure what you expect here. Up to openSUSE 42.1 and
SLE 12 SP1 they are fully maintained, i.e. you can get quite a
bit newer than 3.4 based kernels. There's not a lot of
performance analysis that I would be aware of, so I can't
answer that part anyway. And them being release branches
rather than development ones, there's not going to be any
new feature work.

> Related question is if they have a future (at SUSE) at all? The repo 
> does not have Xen patches for SLE12-SP2. Still coming? Or dropped forever?

Dropped. For the stable-xen branch I'll be maintaining the patches
for a little while, but eventually that'll stop too unless we encounter
a fatal issue on the branches that have got switched to pv-ops.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: state of xenified SUSE kernel

[Andreas Kinzler]

On 17.05.2016 17:34, Jan Beulich wrote:
>> We use xenified kernels based on kernel 3.4 for years and benchmarks
>> showed that they are faster than the pvops (vanilla) kernels.
>> But what is the current state in terms of performance and features?
> I'm not sure what you expect here. Up to openSUSE 42.1 and
> SLE 12 SP1 they are fully maintained, i.e. you can get quite a
> bit newer than 3.4 based kernels. There's not a lot of
> performance analysis that I would be aware of, so I can't
> answer that part anyway. And them being release branches
> rather than development ones, there's not going to be any
> new feature work.

As far as I know the patches are based on the original work for kernel 
2.6.18 and have not been adjusted to major changes in Xen architecture. 
So what I am thinking about is performance improvements that result from 
using newer Xen features. So the question was: from an architectural 
point of view should pvops in recent vanilla kernels be "better" then 
xenified kernels?

Regards Andreas

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: state of xenified SUSE kernel

[Jan Beulich]

>>> On 18.05.16 at 16:39, <ml-ak@posteo.de> wrote:
> On 17.05.2016 17:34, Jan Beulich wrote:
>>> We use xenified kernels based on kernel 3.4 for years and benchmarks
>>> showed that they are faster than the pvops (vanilla) kernels.
>>> But what is the current state in terms of performance and features?
>> I'm not sure what you expect here. Up to openSUSE 42.1 and
>> SLE 12 SP1 they are fully maintained, i.e. you can get quite a
>> bit newer than 3.4 based kernels. There's not a lot of
>> performance analysis that I would be aware of, so I can't
>> answer that part anyway. And them being release branches
>> rather than development ones, there's not going to be any
>> new feature work.
> 
> As far as I know the patches are based on the original work for kernel 
> 2.6.18 and have not been adjusted to major changes in Xen architecture. 
> So what I am thinking about is performance improvements that result from 
> using newer Xen features.

We've added support for newer features where suitable, over the
years. When reasonable I've also tried to put these into the 2.6.18
tree. But from all I can tell right now this isn't going to happen any
further.

> So the question was: from an architectural 
> point of view should pvops in recent vanilla kernels be "better" then 
> xenified kernels?

Hence it's hard to answer this question, the more that I can't even
prove (or disprove) what you've said regarding its performance
having been better in the past.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel