Server product kernel requirements

[Josh Boyer]

[ Resend with the right server mailing list address.  Sorry kernel@ people.]

Hi All,

I realize the WG is just forming up and you have a lot of other items
to cover for now, but I wanted to get this sent out and have people
start thinking about it sooner rather than later.

The kernel team is interested in what the Server WG sees as its
requirements for the kernel package.  Does today's kernel image mostly
suit those needs already, or are there changes that would be
beneficial?

While you think about this, please keep in mind that the kernel team
really wants to keep a single kernel package across all 3 products as
much as possible.  We won't scale to providing multiple kernel
packages or vmlinux binaries for each product.  At the moment, we're
essentially looking for a good "core" kernel package that suits cloud,
server, and workstation and then at repackaging the drivers into
subpackages where appropriate.

If you have changes you'd like to see, please let us know what they
are and the reasoning behind those changes.  Hopefully we can work
with all 3 WGs and come up with something suitable for everyone.
Thanks for your time.

josh

[Simo Sorce]

On Wed, 2013-10-30 at 10:16 -0400, Josh Boyer wrote:
>  [ Resend with the right server mailing list address.  Sorry kernel@
> people.]
>  
>  Hi All,
>  
>  I realize the WG is just forming up and you have a lot of other items
>  to cover for now, but I wanted to get this sent out and have people
>  start thinking about it sooner rather than later.
>  
>  The kernel team is interested in what the Server WG sees as its
>  requirements for the kernel package.  Does today's kernel image mostly
>  suit those needs already, or are there changes that would be
>  beneficial?
>  
>  While you think about this, please keep in mind that the kernel team
>  really wants to keep a single kernel package across all 3 products as
>  much as possible.  We won't scale to providing multiple kernel
>  packages or vmlinux binaries for each product.  At the moment, we're
>  essentially looking for a good "core" kernel package that suits cloud,
>  server, and workstation and then at repackaging the drivers into
>  subpackages where appropriate.
>  
>  If you have changes you'd like to see, please let us know what they
>  are and the reasoning behind those changes.  Hopefully we can work
>  with all 3 WGs and come up with something suitable for everyone.
>  Thanks for your time. 
Personally I think that as long as the kernel is modular and all useful
modules are available, the Server WG should not have trouble with it.

I guess the installation procedure (hence Anaconda) need to be somewhat
customizable so that the server image is by default a lot friendlier to
the type of hardware a server gets to use, and the kind of defaults that
make more sense for a server vs say desktop or cloud.

But I think all this can be built easily above a common kernel.

Simo.

-- 
Simo Sorce * Red Hat, Inc * New York

[Stephen Gallagher]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 10/30/2013 11:10 AM, Simo Sorce wrote:
>  On Wed, 2013-10-30 at 10:16 -0400, Josh Boyer wrote:
> > [ Resend with the right server mailing list address.  Sorry
> > kernel@ people.]
> > 
> > Hi All,
> > 
> > I realize the WG is just forming up and you have a lot of other
> > items to cover for now, but I wanted to get this sent out and
> > have people start thinking about it sooner rather than later.
> > 
> > The kernel team is interested in what the Server WG sees as its 
> > requirements for the kernel package.  Does today's kernel image
> > mostly suit those needs already, or are there changes that would
> > be beneficial?
> > 
> > While you think about this, please keep in mind that the kernel
> > team really wants to keep a single kernel package across all 3
> > products as much as possible.  We won't scale to providing
> > multiple kernel packages or vmlinux binaries for each product.
> > At the moment, we're essentially looking for a good "core" kernel
> > package that suits cloud, server, and workstation and then at
> > repackaging the drivers into subpackages where appropriate.
> > 
> > If you have changes you'd like to see, please let us know what
> > they are and the reasoning behind those changes.  Hopefully we
> > can work with all 3 WGs and come up with something suitable for
> > everyone. Thanks for your time.
>  
>  Personally I think that as long as the kernel is modular and all
>  useful modules are available, the Server WG should not have trouble
>  with it.
>  
>  I guess the installation procedure (hence Anaconda) need to be
>  somewhat customizable so that the server image is by default a lot
>  friendlier to the type of hardware a server gets to use, and the
>  kind of defaults that make more sense for a server vs say desktop
>  or cloud.
>  
>  But I think all this can be built easily above a common kernel.
>   

I agree as well. A common kernel is paramount.

> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.15 (GNU/Linux)
> Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
>
> iEYEARECAAYFAlJxIdAACgkQeiVVYja6o6PrNQCfdMESeLDsmICPezc9i7I/LaDF
> Y44AoJdvRb849P6urc43NjMzDQMou+WG
> =ApQv
> -----END PGP SIGNATURE-----

[Josh Boyer]

On Wed, Oct 30, 2013 at 11:12 AM, Stephen Gallagher <sgallagh(a)redhat.com&gt; wrote:
>  -----BEGIN PGP SIGNED MESSAGE-----
>  Hash: SHA1
>
>  On 10/30/2013 11:10 AM, Simo Sorce wrote:
> > On Wed, 2013-10-30 at 10:16 -0400, Josh Boyer wrote:
> >> [ Resend with the right server mailing list address.  Sorry
> >> kernel@ people.]
> >>
> >> Hi All,
> >>
> >> I realize the WG is just forming up and you have a lot of other
> >> items to cover for now, but I wanted to get this sent out and
> >> have people start thinking about it sooner rather than later.
> >>
> >> The kernel team is interested in what the Server WG sees as its
> >> requirements for the kernel package.  Does today's kernel image
> >> mostly suit those needs already, or are there changes that would
> >> be beneficial?
> >>
> >> While you think about this, please keep in mind that the kernel
> >> team really wants to keep a single kernel package across all 3
> >> products as much as possible.  We won't scale to providing
> >> multiple kernel packages or vmlinux binaries for each product.
> >> At the moment, we're essentially looking for a good "core" kernel
> >> package that suits cloud, server, and workstation and then at
> >> repackaging the drivers into subpackages where appropriate.
> >>
> >> If you have changes you'd like to see, please let us know what
> >> they are and the reasoning behind those changes.  Hopefully we
> >> can work with all 3 WGs and come up with something suitable for
> >> everyone. Thanks for your time.
> >
> > Personally I think that as long as the kernel is modular and all
> > useful modules are available, the Server WG should not have trouble
> > with it.
> >
> > I guess the installation procedure (hence Anaconda) need to be
> > somewhat customizable so that the server image is by default a lot
> > friendlier to the type of hardware a server gets to use, and the
> > kind of defaults that make more sense for a server vs say desktop
> > or cloud.
> >
> > But I think all this can be built easily above a common kernel.
> >
>
>
>  I agree as well. A common kernel is paramount. 
OK, good.  To be fair, I didn't think there would be many requirements
from the Server WG.  Most of the packaging changes will likely be
driven by the Cloud WG.  At the moment, it shouldn't be difficult for
Server to install both the small base kernel package and the larger
kernel-drivers subpackage (for example).

At the same time, it would be good to look over what the current
kernel is providing and see if you notice any glaring omissions in
either drivers or settings.  Or at the very least, perhaps define what
kind of machines you will be targeting with the Server WG spin.
Massive 4096 multi-cored CPU machines with terabytes of DRAM and
petabytes of storage, or more commodity style hardware used in
heterogeneous environments, etc.  Hopefully the impacts to the kernel
there will be minimal, but it would help us understand what you're
shooting for so we can keep it in mind.

josh

[David Timothy Strauss]

On Wed, Oct 30, 2013 at 10:09 AM, Josh Boyer <jwboyer(a)fedoraproject.org&gt; wrote:
>  Massive 4096 multi-cored CPU machines with terabytes of DRAM and
>  petabytes of storage, or more commodity style hardware used in
>  heterogeneous environments, etc. 
The latter. We'd want a separate HPC group for 512+ core machines.

[Simo Sorce]

On Wed, 2013-10-30 at 10:51 -0700, David Strauss wrote:
>  On Wed, Oct 30, 2013 at 10:09 AM, Josh Boyer
> <jwboyer(a)fedoraproject.org&gt; wrote:
>  > Massive 4096 multi-cored CPU machines with terabytes of DRAM and
>  > petabytes of storage, or more commodity style hardware used in
>  > heterogeneous environments, etc.
>  
>  The latter. We'd want a separate HPC group for 512+ core machines. 
Or simply, sites so big can care for their own kernel builds most
probably, or seek for commercial support.


Simo.

-- 
Simo Sorce * Red Hat, Inc * New York

[Jóhann B. Guðmundsson]

On 10/30/2013 05:51 PM, David Strauss wrote:
>  On Wed, Oct 30, 2013 at 10:09 AM, Josh Boyer
> <jwboyer(a)fedoraproject.org&gt; wrote:
> > Massive 4096 multi-cored CPU machines with terabytes of DRAM and
> > petabytes of storage, or more commodity style hardware used in
> > heterogeneous environments, etc.
>  The latter. We'd want a separate HPC group for 512+ core machines. 
I dont see why so please explain?

JBG

[Matthew Miller]

On Wed, Oct 30, 2013 at 06:32:26PM +0000, "Jóhann B. Guðmundsson" wrote:
>  >>Massive 4096 multi-cored CPU machines with terabytes of DRAM
> and
>  >>petabytes of storage, or more commodity style hardware used in
>  >>heterogeneous environments, etc.
>  >The latter. We'd want a separate HPC group for 512+ core machines.
>  I dont see why so please explain? 
HPC is kind of a special case, where performance trumps all else. It would
be pretty awesome to see a Fedora product based around that. It could have
ATLAS packages that dynamically build from source as designed, for one
thing. As heretical as this sounds, SELinux might be off on the nodes, and
so on. 

-- 
Matthew Miller  ☁☁☁  Fedora Cloud Architect  ☁☁☁  <mattdm(a)fedoraproject.org&gt;

[Jóhann B. Guðmundsson]

On 10/30/2013 06:39 PM, Matthew Miller wrote:
>  On Wed, Oct 30, 2013 at 06:32:26PM +0000, "Jóhann B.
> Guðmundsson" wrote:
> >>> Massive 4096 multi-cored CPU machines with terabytes of DRAM and
> >>> petabytes of storage, or more commodity style hardware used in
> >>> heterogeneous environments, etc.
> >> The latter. We'd want a separate HPC group for 512+ core machines.
> > I dont see why so please explain?
>  HPC is kind of a special case, where performance trumps all else. It would
>  be pretty awesome to see a Fedora product based around that. It could have
>  ATLAS packages that dynamically build from source as designed, for one
>  thing. As heretical as this sounds, SELinux might be off on the nodes, and
>  so on.
>  
Still not seeing why that product could not be one of the server 
products or strong argument for it being in it's own group.

JBG

[David Timothy Strauss]

On Wed, Oct 30, 2013 at 11:32 AM, "Jóhann B. Guðmundsson"
<johannbg(a)gmail.com&gt; wrote:
>  On 10/30/2013 05:51 PM, David Strauss wrote:
> > The latter. We'd want a separate HPC group for 512+ core machines.
>  I dont see why so please explain? 
HPC architecture involves atypical services, hardware, and frameworks
versus normal server computing, even compared to thousands of cloud
machines. Wikipedia has some good coverage of popular approaches [1].
A properly built kernel is an essential part of an HPC deployment,
particularly with respect to the scheduler and memory management.
Build options appropriate for embedded all the way to normal servers
are often inappropriate for HPC.

SUSE Linux has a large following in HPC, but it's not just a product
of their general server work.

[1] http://en.wikipedia.org/wiki/Template:Parallel_computing