Xen for Human Beings

classic Classic list List threaded Threaded
22 messages Options
12
Reply | Threaded
Open this post in threaded view
|

Xen for Human Beings

Tom Hibbert-5
The case for Xen virtualisation
-------------------------------

Xen 3.0.2 has opened new opportunities. For the first time, the Xen
kernel is fully modularisable and supports most
hardware. The Xen subarchitecture still has warts - such as not
supporting anything other than a 31 bit mode dma mask,
leaving some hardware like the ice1712 soundcard behind, and not
supporting the nvidia kernel mode driver, though this is likely an
issue to take up with NVIDIA themselves. As one of the targets of Xen
development is inclusion of the subarch in the main Linux kernel
distribution, we can expect to see these problems cleared up in the near
future.

Xen has proved extremely useful in the server environment, as well as on
 desktops for Linux development.

Testing Ubuntu livecds can be tedious, because sometimes they take an
annoyingly long time to start up because of all
the autodetection and questions asked.
Granted, this does make them work really well once they've booted.
Autodetection of almost almost hardware is seamless.
However this process must be performed many times and can get tedious.
Hence the idea of distributing Xen images of the liveCDs that people can
boot up on their Dapper workstations to play with.

Steps in the right direction
----------------------------

I've proven that Ubuntu is very usable under Xen 3.0.2 right now, and is
genuinely useful as a dev environment - I can
get my hands dirty in etch on a scratch VM in a matter of seconds.

My packages are built from the Debian repository with really very little
changes. They could be synced into the
universe repository or maintained in a semi-official manner at an
external site.
(http://debian.thoughtcrime.org.nz/ubuntu/)

Building the xenlinux kernel should perhaps be left up to the user. i
have some notes that will be wikified probably
later today.

Xenlinux kernels built against ubuntu kernels do boot, albeit with less
hardware support then the standard kernel.
I havent begun to explore the Ubuntu specific patch sets.

I would like to work closely with the kernel team to build a usable kernel
that works on a wide range of systems including workstations, so it can
be used to test bleeding edge Ubuntu development
without having to sacrifice the stability of your workstation.


Xen also has tremendous power at the server end. It offers huge
flexibility to the administrator with regard to service
deployment and scalability, as well as providing virtual staging grounds
for administrators to easily test operational
changes to production server blocs. It can be leveraged to greatly
streamline the development and deployment
of the Ubuntu datacenter.



The Xenubuntu project
---------------------


The Xenubuntu project would begin by remastering the lightweight xubuntu
livecd into a liveDVD containing casper images of all the *buntus,
as well as state files of them fully booted up (~256mb ram, if
practical), that boots into a GUI that can connect via
VNC/xDMCP/FreeNX/fooprotocol to the guest and displays it in full screen.

>From there we can find more fun stuff to do with Xen and Ubuntu.

I tried to edit the xen kernels and Xen and NX integration specs on
launchpad, but apparently i dont have permission to do so. What do I
need to do to be able to edit these?

Thanks

Tom

--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Dennis Kaarsemaker
On di, 2006-04-25 at 12:24 +1200, Tom Hibbert wrote:
> I tried to edit the xen kernels and Xen and NX integration specs on
> launchpad, but apparently i dont have permission to do so. What do I
> need to do to be able to edit these?

Edit the wikipage they point to :)
--
Dennis K.
 - Linux for human beings - http://www.ubuntu.com
 - Linux voor iedereen - http://www.ubuntu-nl.org

--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel

signature.asc (198 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Raphaël Pinson
In reply to this post by Tom Hibbert-5
Hi Tom,

If you plan to include these packages in Ubuntu, you should get them on REVU
(http://revu.tauware.de). Please read
https://wiki.ubuntu.com/MOTU/Packages/REVU on the wiki for further
informations.

Cheers,


Raphaël


Le Mardi 25 Avril 2006 02:24, Tom Hibbert a écrit :

> The case for Xen virtualisation
> -------------------------------
>
> Xen 3.0.2 has opened new opportunities. For the first time, the Xen
> kernel is fully modularisable and supports most
> hardware. The Xen subarchitecture still has warts - such as not
> supporting anything other than a 31 bit mode dma mask,
> leaving some hardware like the ice1712 soundcard behind, and not
> supporting the nvidia kernel mode driver, though this is likely an
> issue to take up with NVIDIA themselves. As one of the targets of Xen
> development is inclusion of the subarch in the main Linux kernel
> distribution, we can expect to see these problems cleared up in the near
> future.
>
> Xen has proved extremely useful in the server environment, as well as on
>  desktops for Linux development.
>
> Testing Ubuntu livecds can be tedious, because sometimes they take an
> annoyingly long time to start up because of all
> the autodetection and questions asked.
> Granted, this does make them work really well once they've booted.
> Autodetection of almost almost hardware is seamless.
> However this process must be performed many times and can get tedious.
> Hence the idea of distributing Xen images of the liveCDs that people can
> boot up on their Dapper workstations to play with.
>
> Steps in the right direction
> ----------------------------
>
> I've proven that Ubuntu is very usable under Xen 3.0.2 right now, and is
> genuinely useful as a dev environment - I can
> get my hands dirty in etch on a scratch VM in a matter of seconds.
>
> My packages are built from the Debian repository with really very little
> changes. They could be synced into the
> universe repository or maintained in a semi-official manner at an
> external site.
> (http://debian.thoughtcrime.org.nz/ubuntu/)
>
> Building the xenlinux kernel should perhaps be left up to the user. i
> have some notes that will be wikified probably
> later today.
>
> Xenlinux kernels built against ubuntu kernels do boot, albeit with less
> hardware support then the standard kernel.
> I havent begun to explore the Ubuntu specific patch sets.
>
> I would like to work closely with the kernel team to build a usable kernel
> that works on a wide range of systems including workstations, so it can
> be used to test bleeding edge Ubuntu development
> without having to sacrifice the stability of your workstation.
>
>
> Xen also has tremendous power at the server end. It offers huge
> flexibility to the administrator with regard to service
> deployment and scalability, as well as providing virtual staging grounds
> for administrators to easily test operational
> changes to production server blocs. It can be leveraged to greatly
> streamline the development and deployment
> of the Ubuntu datacenter.
>
>
>
> The Xenubuntu project
> ---------------------
>
>
> The Xenubuntu project would begin by remastering the lightweight xubuntu
> livecd into a liveDVD containing casper images of all the *buntus,
> as well as state files of them fully booted up (~256mb ram, if
> practical), that boots into a GUI that can connect via
> VNC/xDMCP/FreeNX/fooprotocol to the guest and displays it in full screen.
>
> >From there we can find more fun stuff to do with Xen and Ubuntu.
>
> I tried to edit the xen kernels and Xen and NX integration specs on
> launchpad, but apparently i dont have permission to do so. What do I
> need to do to be able to edit these?
>
> Thanks
>
> Tom
--
Raphaël Pinson
<[hidden email]>
Ubuntu - Linux for Human Beings
http://www.ubuntulinux.org

--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel

attachment0 (198 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Reinhard Tartler-2
In reply to this post by Tom Hibbert-5
On Tue, Apr 25, 2006 at 12:24:08PM +1200, Tom Hibbert wrote:
> My packages are built from the Debian repository with really very little
> changes.

I looked into your packages some time ago, and noticed that they are
heavily based on the xen-3.0 source package, which finally entered
debian unstable. The only changed needed so far was the location of the
udev script, but else the package was working fine for me. I'd support
inclusion of a xen-3.0 package in dapper/universe, requiring users to
run custom kernels. It's universe, after all.

> Building the xenlinux kernel should perhaps be left up to the user. i
> have some notes that will be wikified probably later today.

Care to tell us the URL to the wiki page?

> Xenlinux kernels built against ubuntu kernels do boot, albeit with less
> hardware support then the standard kernel.
> I havent begun to explore the Ubuntu specific patch sets.
> I would like to work closely with the kernel team to build a usable kernel
> that works on a wide range of systems including workstations, so it can
> be used to test bleeding edge Ubuntu development
> without having to sacrifice the stability of your workstation.

This would be really  nice, but this is really a edgy thing. Don't
expect this to enter dapper.

> The Xenubuntu project
> ---------------------
>
> The Xenubuntu project would begin by remastering the lightweight xubuntu
> livecd into a liveDVD containing casper images of all the *buntus,
> as well as state files of them fully booted up (~256mb ram, if
> practical), that boots into a GUI that can connect via
> VNC/xDMCP/FreeNX/fooprotocol to the guest and displays it in full screen.
>
> >From there we can find more fun stuff to do with Xen and Ubuntu.
>
> I tried to edit the xen kernels and Xen and NX integration specs on
> launchpad, but apparently i dont have permission to do so. What do I
> need to do to be able to edit these?

launchpad specific question are more likely to be answered on the
launchpad-users mailing list. Please reask there.

Where can I learn more about the xenubuntu project? Do you have a
Project page or something?

Gruesse,
        Reinhard
 

--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Hein-Pieter van Braam
In reply to this post by Tom Hibbert-5
I have already done a merger of ubuntu patchset (of breezy) and the xen
patches some time ago, the fruits of this work are available at:

http://81.71.31.44/~hp/xen/

perhaps it is of some use to you :) consult the HOWTO in that directory
on what I did to it.

good plan btw! Have you looked at solaris zones? I think it would be a
good way to model the xen virtual machines after them.

have fun! :)

On Tue, 2006-04-25 at 12:24 +1200, Tom Hibbert wrote:

> The case for Xen virtualisation
> -------------------------------
>
> Xen 3.0.2 has opened new opportunities. For the first time, the Xen
> kernel is fully modularisable and supports most
> hardware. The Xen subarchitecture still has warts - such as not
> supporting anything other than a 31 bit mode dma mask,
> leaving some hardware like the ice1712 soundcard behind, and not
> supporting the nvidia kernel mode driver, though this is likely an
> issue to take up with NVIDIA themselves. As one of the targets of Xen
> development is inclusion of the subarch in the main Linux kernel
> distribution, we can expect to see these problems cleared up in the near
> future.
>
> Xen has proved extremely useful in the server environment, as well as on
>  desktops for Linux development.
>
> Testing Ubuntu livecds can be tedious, because sometimes they take an
> annoyingly long time to start up because of all
> the autodetection and questions asked.
> Granted, this does make them work really well once they've booted.
> Autodetection of almost almost hardware is seamless.
> However this process must be performed many times and can get tedious.
> Hence the idea of distributing Xen images of the liveCDs that people can
> boot up on their Dapper workstations to play with.
>
> Steps in the right direction
> ----------------------------
>
> I've proven that Ubuntu is very usable under Xen 3.0.2 right now, and is
> genuinely useful as a dev environment - I can
> get my hands dirty in etch on a scratch VM in a matter of seconds.
>
> My packages are built from the Debian repository with really very little
> changes. They could be synced into the
> universe repository or maintained in a semi-official manner at an
> external site.
> (http://debian.thoughtcrime.org.nz/ubuntu/)
>
> Building the xenlinux kernel should perhaps be left up to the user. i
> have some notes that will be wikified probably
> later today.
>
> Xenlinux kernels built against ubuntu kernels do boot, albeit with less
> hardware support then the standard kernel.
> I havent begun to explore the Ubuntu specific patch sets.
>
> I would like to work closely with the kernel team to build a usable kernel
> that works on a wide range of systems including workstations, so it can
> be used to test bleeding edge Ubuntu development
> without having to sacrifice the stability of your workstation.
>
>
> Xen also has tremendous power at the server end. It offers huge
> flexibility to the administrator with regard to service
> deployment and scalability, as well as providing virtual staging grounds
> for administrators to easily test operational
> changes to production server blocs. It can be leveraged to greatly
> streamline the development and deployment
> of the Ubuntu datacenter.
>
>
>
> The Xenubuntu project
> ---------------------
>
>
> The Xenubuntu project would begin by remastering the lightweight xubuntu
> livecd into a liveDVD containing casper images of all the *buntus,
> as well as state files of them fully booted up (~256mb ram, if
> practical), that boots into a GUI that can connect via
> VNC/xDMCP/FreeNX/fooprotocol to the guest and displays it in full screen.
>
> >From there we can find more fun stuff to do with Xen and Ubuntu.
>
> I tried to edit the xen kernels and Xen and NX integration specs on
> launchpad, but apparently i dont have permission to do so. What do I
> need to do to be able to edit these?
>
> Thanks
>
> Tom
>


--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Erast Benson
> good plan btw! Have you looked at solaris zones? I think it would be a
> good way to model the xen virtual machines after them.
>
> have fun! :)

Yes, Solaris 10 zones are nice but limited in the way that it can only
"run" another instance of the same version of Solaris. It is useful to
some extent and runs without *any* performance loss.

But upcoming BrandZ[1] (Solaris 11 zones) will have additional syscall
layer which will allow to run OpenSolaris-based distribution like
NexentaOS[2] without *any* performance loss and Linux-based distributions
with minimal performance loss (5-10%), which is way more efficient than
Xen/VMware/etc..

BrandZ patches are freely available now and we are planning to include
them into the NexentaOS once they are part of an official build.

[1] http://www.opensolaris.org/os/community/brandz
[2] http://www.gnusolaris.org


--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Hein-Pieter van Braam
Actually, I meant the way you create and manage them :)
Not on a technical level really :)

On Wed, 2006-04-26 at 10:36 -0700, Erast Benson wrote:

> > good plan btw! Have you looked at solaris zones? I think it would be a
> > good way to model the xen virtual machines after them.
> >
> > have fun! :)
>
> Yes, Solaris 10 zones are nice but limited in the way that it can only
> "run" another instance of the same version of Solaris. It is useful to
> some extent and runs without *any* performance loss.
>
> But upcoming BrandZ[1] (Solaris 11 zones) will have additional syscall
> layer which will allow to run OpenSolaris-based distribution like
> NexentaOS[2] without *any* performance loss and Linux-based distributions
> with minimal performance loss (5-10%), which is way more efficient than
> Xen/VMware/etc..
>
> BrandZ patches are freely available now and we are planning to include
> them into the NexentaOS once they are part of an official build.
>
> [1] http://www.opensolaris.org/os/community/brandz
> [2] http://www.gnusolaris.org
>
>


--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Ivan Krstić-3
In reply to this post by Erast Benson
Erast Benson wrote:
> But upcoming BrandZ[1] (Solaris 11 zones) will have additional syscall
> layer which will allow to run OpenSolaris-based distribution like
> NexentaOS[2] without *any* performance loss

That's exceedingly unlikely. Every layer of abstraction that moves you
away from bare metal necessarily introduces some performance degradation.

> and Linux-based distributions
> with minimal performance loss (5-10%), which is way more efficient than
> Xen/VMware/etc..

Please read up on Xen's performance before making factually incorrect
claims.

--
Ivan Krstic <[hidden email]> | GPG: 0x147C722D

--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Santiago Roza-2
In reply to this post by Erast Benson
On 4/26/06, Erast Benson <[hidden email]> wrote:

> ... and Linux-based distributions
> with minimal performance loss (5-10%), which is way more efficient than
> Xen/VMware/etc..


vmware and xen don't use the same techniques, and vmware's performance
overhead is considerably higher:
http://en.wikipedia.org/wiki/Xen_(virtual_machine_monitor)#Comparison_with_other_Virtual_Machine_Managers_.28VMMs.29


--
Santiago Roza
Departamento I+D - Thymbra
[hidden email]

--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Erast Benson
In reply to this post by Ivan Krstić-3
> Erast Benson wrote:
>> But upcoming BrandZ[1] (Solaris 11 zones) will have additional syscall
>> layer which will allow to run OpenSolaris-based distribution like
>> NexentaOS[2] without *any* performance loss
>
> That's exceedingly unlikely. Every layer of abstraction that moves you
> away from bare metal necessarily introduces some performance degradation.

Please read what BrandZ is. There is no HW abstraction introduced. It only
introduces tiny software layer of syscall's abstraction. Look at  "3.4.
System Call Emulation" chapter at Design document[1].

>> and Linux-based distributions
>> with minimal performance loss (5-10%), which is way more efficient than
>> Xen/VMware/etc..
>
> Please read up on Xen's performance before making factually incorrect
> claims.

Nope. Xen technology badly loosing performance game to Zones, especially
when you start running 10-20 nodes. Zones are native. You could think of
Zone as of chroot environment with networking interface. Do you ever seen
noticable performance regression in chroot environment?

Having said that OpenSolaris's Xen port is ongoing effort[2] and I think
it is pretty usable already. And yes, Xen has its own benefits, like
hypervisor which overtime will benefit from Ring -1 of new generation
Intel/AMD CPU(s). But it will take time till all this hardware will become
common (yes motherboards, PCI bridges, etc needs to be changed too...)

[1] http://www.opensolaris.org/os/community/brandz/design
[2] http://www.opensolaris.org/os/community/xen


--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Erast Benson
In reply to this post by Santiago Roza-2
> On 4/26/06, Erast Benson <[hidden email]> wrote:
>
>> ... and Linux-based distributions
>> with minimal performance loss (5-10%), which is way more efficient than
>> Xen/VMware/etc..
>
>
> vmware and xen don't use the same techniques, and vmware's performance
> overhead is considerably higher:
> http://en.wikipedia.org/wiki/Xen_(virtual_machine_monitor)#Comparison_with_other_Virtual_Machine_Managers_.28VMMs.29

Correct. Xen != UML != VMware != BrandZ

I tried them all and I would sort them in this way (top - best performance):

BrandZ
Xen
UML
VMware


--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Matthew Palmer-2
On Wed, Apr 26, 2006 at 11:59:46AM -0700, Erast Benson wrote:

> > On 4/26/06, Erast Benson <[hidden email]> wrote:
> >
> >> ... and Linux-based distributions
> >> with minimal performance loss (5-10%), which is way more efficient than
> >> Xen/VMware/etc..
> >
> >
> > vmware and xen don't use the same techniques, and vmware's performance
> > overhead is considerably higher:
> > http://en.wikipedia.org/wiki/Xen_(virtual_machine_monitor)#Comparison_with_other_Virtual_Machine_Managers_.28VMMs.29
>
> Correct. Xen != UML != VMware != BrandZ
>
> I tried them all and I would sort them in this way (top - best performance):
>
> BrandZ
> Xen

What is the quantitative difference in performance between these two?  What
number is large enough for you to rate BrandZ as "way more efficient"?

- Matt

--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Matthew Palmer-2
In reply to this post by Erast Benson
On Wed, Apr 26, 2006 at 11:40:50AM -0700, Erast Benson wrote:

> > Erast Benson wrote:
> >> But upcoming BrandZ[1] (Solaris 11 zones) will have additional syscall
> >> layer which will allow to run OpenSolaris-based distribution like
> >> NexentaOS[2] without *any* performance loss
> >
> > That's exceedingly unlikely. Every layer of abstraction that moves you
> > away from bare metal necessarily introduces some performance degradation.
>
> Please read what BrandZ is. There is no HW abstraction introduced. It only
> introduces tiny software layer of syscall's abstraction. Look at  "3.4.

And this tiny software layer manages to do it's work in zero time?  Amazing.
They should use this marvellous technology for the rest of the OS, and then
it can do all possible work in zero time.

Or maybe there *is* a performance loss after all.

> >> and Linux-based distributions
> >> with minimal performance loss (5-10%), which is way more efficient than
> >> Xen/VMware/etc..
> >
> > Please read up on Xen's performance before making factually incorrect
> > claims.
>
> Nope. Xen technology badly loosing performance game to Zones, especially
> when you start running 10-20 nodes.

Show us the numbers.

> Zones are native. You could think of Zone as of chroot environment with
> networking interface.

So like Linux vservers then, just without the process segregation and
CPU/memory limits?

- Matt

--
I'm personally quite happy with one stable release every two years, and am
of the opinion that trying to release more will mean we'll have to rename
the distro from "stable" to "wobbly". -- Scott Remnant on debian-devel

--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Erast Benson
In reply to this post by Matthew Palmer-2
On Thu, 2006-04-27 at 07:59 +1000, Matthew Palmer wrote:

> On Wed, Apr 26, 2006 at 11:59:46AM -0700, Erast Benson wrote:
> > > On 4/26/06, Erast Benson <[hidden email]> wrote:
> > >
> > >> ... and Linux-based distributions
> > >> with minimal performance loss (5-10%), which is way more efficient than
> > >> Xen/VMware/etc..
> > >
> > >
> > > vmware and xen don't use the same techniques, and vmware's performance
> > > overhead is considerably higher:
> > > http://en.wikipedia.org/wiki/Xen_(virtual_machine_monitor)#Comparison_with_other_Virtual_Machine_Managers_.28VMMs.29
> >
> > Correct. Xen != UML != VMware != BrandZ
> >
> > I tried them all and I would sort them in this way (top - best performance):
> >
> > BrandZ
> > Xen
>
> What is the quantitative difference in performance between these two?  What
> number is large enough for you to rate BrandZ as "way more efficient"?

It runs with 0% overhead. Means *significantly* faster. Name your
number... 2x, 5x, 10x times faster will work too! :-)

Here is some more elaboration from the Solaris 10 Zone FAQ[1]:

"""What's the Relationship between Xen and Zones?

Some people think of Xen and Zones as competing alternatives, when in
reality they are used in a complementary fashion. That's because the two
technologies operate at different levels in the software stack. Xen
virtualizes the hardware, which allows multiple operating system kernels
to run simultaneously on (currently) one piece of physical hardware.
Solaris Zones virtualize the OS, allowing multiple operating system
environments to coexist on a single Solaris kernel. Thus the two
technologies can either be used independently or used in combination to
provide whatever set of attributes best fit the problem at hand.

The different approaches possess efficiency, observability, sharing,
utilization, isolation and manageability attributes which can be
markedly different. The Xen approach is good for isolation, though
presents the need to manage several complete operating environments,
which can be a feature or a problem, depending on what you're trying to
do! In contrast, because of the high degree of resource sharing that is
possible, Zones are good for utilization, observability, and
manageability. """

And now about BrandZ (which is logical extension of Zones)[2]:

""" What is the difference between BrandZ and Xen?

Xen is a technology that virtualizes an entire machine, allowing you to
run a complete operating system including the kernel. BrandZ allows you
to run just an operating system's userspace components, with a Solaris
kernel under the hood. """

Once BrandZ is out, I will try to wrap it in into the NexentaOS, so
people will have a chance to try it out in familiar Ubuntu-based
environment.

[1] http://www.opensolaris.org/os/community/xen/solaris-on-xen-faq
[2] http://www.opensolaris.org/os/community/brandz/brandz_lae_faq

--
Erast


--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Matthew Palmer-2
In reply to this post by Erast Benson
On Wed, Apr 26, 2006 at 10:36:46AM -0700, Erast Benson wrote:
> But upcoming BrandZ[1] (Solaris 11 zones) will have additional syscall
> layer which will allow to run OpenSolaris-based distribution like
> NexentaOS[2] without *any* performance loss and Linux-based distributions
> with minimal performance loss (5-10%), which is way more efficient than
> Xen/VMware/etc..

5% overhead is about what Xen takes for paravirtualised domains.  From the
referenced page, I don't see any mention of full virtualisation, so you
won't be able to run other OSes at all in BrandZ (who thought up that name,
anyway?), either.

- Matt

--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Matthew East
In reply to this post by Matthew Palmer-2
On Thu, 2006-04-27 at 08:36 +1000, Matthew Palmer wrote:
> On Wed, Apr 26, 2006 at 11:40:50AM -0700, Erast Benson wrote:
> > > Erast Benson wrote:
> > >> But upcoming BrandZ[1] (Solaris 11 zones) will have additional syscall
> > >> layer which will allow to run OpenSolaris-based distribution like
> > >> NexentaOS[2] without *any* performance loss

> Show us the numbers.

But please do it on an appropriate mailing list: sounder would be good.
I'm including sounder in this message, so that you can just delete
-devel when next replyingl

Thanks!

Matt
--
[hidden email]
gnupg pub 1024D/0E6B06FF

--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel

signature.asc (198 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

andrewski
In reply to this post by Matthew Palmer-2
I think this is no longer -devel discussion.

On 4/26/06, Matthew Palmer <[hidden email]> wrote:

> On Wed, Apr 26, 2006 at 11:40:50AM -0700, Erast Benson wrote:
> > > Erast Benson wrote:
> > >> But upcoming BrandZ[1] (Solaris 11 zones) will have additional syscall
> > >> layer which will allow to run OpenSolaris-based distribution like
> > >> NexentaOS[2] without *any* performance loss
> > >
> > > That's exceedingly unlikely. Every layer of abstraction that moves you
> > > away from bare metal necessarily introduces some performance degradation.
> >
> > Please read what BrandZ is. There is no HW abstraction introduced. It only
> > introduces tiny software layer of syscall's abstraction. Look at  "3.4.
>
> And this tiny software layer manages to do it's work in zero time?  Amazing.
> They should use this marvellous technology for the rest of the OS, and then
> it can do all possible work in zero time.
>
> Or maybe there *is* a performance loss after all.
>
> > >> and Linux-based distributions
> > >> with minimal performance loss (5-10%), which is way more efficient than
> > >> Xen/VMware/etc..
> > >
> > > Please read up on Xen's performance before making factually incorrect
> > > claims.
> >
> > Nope. Xen technology badly loosing performance game to Zones, especially
> > when you start running 10-20 nodes.
>
> Show us the numbers.
>
> > Zones are native. You could think of Zone as of chroot environment with
> > networking interface.
>
> So like Linux vservers then, just without the process segregation and
> CPU/memory limits?
>
> - Matt
>
> --
> I'm personally quite happy with one stable release every two years, and am
> of the opinion that trying to release more will mean we'll have to rename
> the distro from "stable" to "wobbly". -- Scott Remnant on debian-devel
>
> --
> ubuntu-devel mailing list
> [hidden email]
> https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
>

--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Erast Benson
In reply to this post by Matthew Palmer-2
On Thu, 2006-04-27 at 08:39 +1000, Matthew Palmer wrote:

> On Wed, Apr 26, 2006 at 10:36:46AM -0700, Erast Benson wrote:
> > But upcoming BrandZ[1] (Solaris 11 zones) will have additional syscall
> > layer which will allow to run OpenSolaris-based distribution like
> > NexentaOS[2] without *any* performance loss and Linux-based distributions
> > with minimal performance loss (5-10%), which is way more efficient than
> > Xen/VMware/etc..
>
> 5% overhead is about what Xen takes for paravirtualised domains.  From the
> referenced page, I don't see any mention of full virtualisation, so you
> won't be able to run other OSes at all in BrandZ (who thought up that name,
> anyway?), either.

Sure, 5% is the overhead for an "alien" Branded Zone. Only Linux
available at the moment. Native Zone do not introduce any visible
overhead.

--
Erast


--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Tom Hibbert-5
In reply to this post by Reinhard Tartler-2
Reinhard Tartler wrote:
> I looked into your packages some time ago, and noticed that they are
> heavily based on the xen-3.0 source package, which finally entered
> debian unstable. The only changed needed so far was the location of the
> udev script, but else the package was working fine for me. I'd support
> inclusion of a xen-3.0 package in dapper/universe, requiring users to
> run custom kernels. It's universe, after all.

Thats the way I'd like to see it go - working with the Debian xen
maintainers.

>> Building the xenlinux kernel should perhaps be left up to the user. i
>> have some notes that will be wikified probably later today.
>
> Care to tell us the URL to the wiki page?

Sure. https://wiki.ubuntu.com/XenVirtualMachine/XenOnUbuntuDapper

Note that due to 40088 still not being merged that this isnt the Right Way.

>> Xenlinux kernels built against ubuntu kernels do boot, albeit with less
>> hardware support then the standard kernel.
>> I havent begun to explore the Ubuntu specific patch sets.
>> I would like to work closely with the kernel team to build a usable kernel
>> that works on a wide range of systems including workstations, so it can
>> be used to test bleeding edge Ubuntu development
>> without having to sacrifice the stability of your workstation.
>
> This would be really  nice, but this is really a edgy thing. Don't
> expect this to enter dapper.

Agreed.

>> I tried to edit the xen kernels and Xen and NX integration specs on
>> launchpad, but apparently i dont have permission to do so. What do I
>> need to do to be able to edit these?
>
> launchpad specific question are more likely to be answered on the
> launchpad-users mailing list. Please reask there.

Sorted now. I've drafted the spec for xen-enabled-kernel, you may read
it here: https://wiki.ubuntu.com/XenEnabledKernel


> Where can I learn more about the xenubuntu project? Do you have a
> Project page or something?

Not yet although I have added it to blueprint. I dont think it would
take me very long to do either, just need to get the time to do it.

Thanks

Tom

--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
Reply | Threaded
Open this post in threaded view
|

Re: Xen for Human Beings

Reinhard Tartler-2
In reply to this post by Erast Benson
I'm CC'ing [hidden email], let's please continue this
thread there. This is getting more and more into a chatter than
development discussion.

On Wed, Apr 26, 2006 at 10:36:46AM -0700, Erast Benson wrote:
> > good plan btw! Have you looked at solaris zones? I think it would be a
> > good way to model the xen virtual machines after them.
> >
> > have fun! :)
>
> Yes, Solaris 10 zones are nice but limited in the way that it can only
> "run" another instance of the same version of Solaris. It is useful to
> some extent and runs without *any* performance loss.

Sounds very similar to what linux vserver patches provide. [1] The kernels
in debian/unstable seem to already provide such a kernel variant.
 
> But upcoming BrandZ[1] (Solaris 11 zones) will have additional syscall
> layer which will allow to run OpenSolaris-based distribution like
> NexentaOS[2] without *any* performance loss and Linux-based distributions
> with minimal performance loss (5-10%), which is way more efficient than
> Xen/VMware/etc..
>
> BrandZ patches are freely available now and we are planning to include
> them into the NexentaOS once they are part of an official build.

Well, from what I understand, they offer (on i386 only) some syscall,
to enable x86 binaries to run on a solaris kernel. Right? Does this mean
you can debootstrap ubuntu 5.10 or ubuntu 6.06 in a solaris brand and
use all functionality? What about tools like iptables, tc and iproute2?

Gruesse,
        Reinhard

[1] http://linux-vserver.org/

--
ubuntu-devel mailing list
[hidden email]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel
12