Xen users - Apr 2007 - Is Xen only in guests slow (tls.disabled)

Hello xen freaks

I have a question. Is xen only in xen guests slow or also in dom0
when I do not execute

mv /lib/tls /lib/tls.disabled?

Thanks for information.
Marco


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

> I have a question. Is xen only in xen guests slow or also in dom0
> when I do not execute
> 
> mv /lib/tls /lib/tls.disabled?
Apps will be slow to run in any domain where you don''t do this,
including dom0.

The exception is HVM (fully virtualised) domains, which don''t need this
to be done.

Cheers,
Mark


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Hi Mark

Mark Williamson schrieb:
> Apps will be slow to run in any domain where you don''t do this,
> including dom0.
Thank you for information.
Marco

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Marco Balmer wrote:
> Hi Mark
>
> Mark Williamson schrieb:
>   
>> Apps will be slow to run in any domain where you don''t do
this,
>> including dom0.
>>     
>
> Thank you for information.
> Marco
>   
There *is* a glibc patch to address this, and RPM''s published for it at
www.xensource.net. The patch is quite small: perhaps you could take a 
look and incorporate it for your use?

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Hi Nico

Is there also a Debian package available?

Marco

Nico Kadel-Garcia schrieb:
> There *is* a glibc patch to address this, and RPM''s published for
it at
> www.xensource.net. The patch is quite small: perhaps you could take a
> look and incorporate it for your use?
_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Marco Balmer wrote:
> Hi Nico
>
> Is there also a Debian package available?
>
> Marco
>
> Nico Kadel-Garcia schrieb:
>   
>> There *is* a glibc patch to address this, and RPM''s published
for it at
>> www.xensource.net. The patch is quite small: perhaps you could take a
>> look and incorporate it for your use?
>>     
I haven''t looked there: Xensource does not publish .deb files. But if 
you can take apart a SRPM, it''s a very modest patch.

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

On Tue, Apr 10, 2007 at 06:54:32PM +0100, Mark Williamson
wrote:
> > when I do not execute
> > 
> > mv /lib/tls /lib/tls.disabled?
> The exception is HVM (fully virtualised) domains, which don''t need
this
> to be done.
  My understanding was that 64-bit dom0/domU were another exception ..

Steve
-- 
# Commercial Debian GNU/Linux Support
http://www.linux-administration.org/


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

> > The exception is HVM (fully virtualised) domains, which don''t
need this
> > to be done.
> 
>   My understanding was that 64-bit dom0/domU were another exception ..
Quite right Steve, sorry for forgetting this - my head spends most of
its time in 32-bit land :-)

Also, Xen/IA64 and Xen/PPC do not have this issue AFAIK.

I don''t know if it''s an issue for 32-bit PAE guests running on
a 64-bit
hypervisor, which will be allowed by 3.0.5...

Cheers,
Mark


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Hi Mark

Is Xen 32-bit non-pae not affected of the slow gests problem?
Sorry a lot of questions, I am xen newbi...

Marco
> Quite right Steve, sorry for forgetting this - my head spends most of
> its time in 32-bit land :-)
> Also, Xen/IA64 and Xen/PPC do not have this issue AFAIK.
> I don''t know if it''s an issue for 32-bit PAE guests
running on a 64-bit
> hypervisor, which will be allowed by 3.0.5...


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

On 11/4/07 00:53, "Mark Williamson" <maw48@cam.ac.uk> wrote:
> Quite right Steve, sorry for forgetting this - my head spends most of
> its time in 32-bit land :-)
> 
> Also, Xen/IA64 and Xen/PPC do not have this issue AFAIK.
> 
> I don''t know if it''s an issue for 32-bit PAE guests
running on a 64-bit
> hypervisor, which will be allowed by 3.0.5...
PAE guests on 64-bit hypervisor will not have this issue.

 -- Keir



_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Hi Marco,

http://packages.debian.org/stable/libs/libc6-xen
> Hi Nico
>
> Is there also a Debian package available?
>
> Marco
>
> Nico Kadel-Garcia schrieb:
>> There *is* a glibc patch to address this, and RPM''s published
for it at
>> www.xensource.net. The patch is quite small: perhaps you could take a
>> look and incorporate it for your use?
>
> _______________________________________________
> Xen-users mailing list
> Xen-users@lists.xensource.com
> http://lists.xensource.com/xen-users
>

Un saludote.

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

> > I don''t know if it''s an issue for 32-bit PAE guests
running on a 64-bit
> > hypervisor, which will be allowed by 3.0.5...
>
> PAE guests on 64-bit hypervisor will not have this issue.
Thanks for clearing that up Keir, I suspected that would be the case.

I''ve not looked at the 32-on-64 code at all, but I''ve been
assuming it would
do something scary like protecting Xen by placing it outside the 32-bit 
address space (with side benefit that system calls in the guests have less 
context switching overhead relative to 64-on-64).  And that neatly fixes the 
wrap-around segmentation problem too.

So I guess that would mean that 64-bit chips are well suited for running 
32-bit domains...  and 32-bit domains are ideally suited to running on 64-bit 
Xen.  Crazy, but cool :-)

Cheers,
Mark

-- 
Dave: Just a question. What use is a unicyle with no seat?  And no pedals!
Mark: To answer a question with a question: What use is a skateboard?
Dave: Skateboards have wheels.
Mark: My wheel has a wheel!

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

> Is Xen 32-bit non-pae not affected of the slow gests problem?
> Sorry a lot of questions, I am xen newbi...
Non-PAE is also affected: any 32-bit paravirtualised Linux on an x86_32 
hypervisor will be affected.

64-bit fixes the problem for both 64-bit and 32-bit paravirt guests.  Other 
architectures didn''t have this problem in the first place.

It originally occurred because 32-bit x86 Xen and 32-bit x86 glibc both do 
weird things with segmentation - and they don''t coexist well as a
result.
SuSE just remove the segmentation trick from glibc, as it has negligible 
performance impact.

Alternative packages are available for other distros.  I think RedHat''s
glibc
might be able to switch modes for optimum performance on Xen or bare metal, 
but I may be misremembering...

Cheers,
Mark

-- 
Dave: Just a question. What use is a unicyle with no seat?  And no pedals!
Mark: To answer a question with a question: What use is a skateboard?
Dave: Skateboards have wheels.
Mark: My wheel has a wheel!

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

On Wed, Apr 11, 2007 at 07:09:53PM +0100, Mark Williamson
wrote:
> > Is Xen 32-bit non-pae not affected of the slow gests problem?
> > Sorry a lot of questions, I am xen newbi...
> 
> Non-PAE is also affected: any 32-bit paravirtualised Linux on an x86_32 
> hypervisor will be affected.
> 
> 64-bit fixes the problem for both 64-bit and 32-bit paravirt guests.  Other
> architectures didn''t have this problem in the first place.
> 
> It originally occurred because 32-bit x86 Xen and 32-bit x86 glibc both do 
> weird things with segmentation - and they don''t coexist well as a
result.
> SuSE just remove the segmentation trick from glibc, as it has negligible 
> performance impact.
> 
> Alternative packages are available for other distros.  I think
RedHat''s glibc
> might be able to switch modes for optimum performance on Xen or bare metal,
> but I may be misremembering...
Our  glibc  RPM installs two different builds on 32-bit - one in /lib and
a second in /lib/i686/nosegneg.  The ELF dynamic loader can automatically
detect the correct version to use based on kernel capabilities. So if
you''re
running bare metal it''ll use the default glibc, and if you then reboot
into
Xen it''ll transparently switch to the nosegneg version. That way both
kernels
get to use the optimum version of glibc.

Regards,
Dan.
-- 
|=- Red Hat, Engineering, Emerging Technologies, Boston.  +1 978 392 2496 -=|
|=-           Perl modules: http://search.cpan.org/~danberr/              -=|
|=-               Projects: http://freshmeat.net/~danielpb/               -=|
|=-  GnuPG: 7D3B9505   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505  -=| 

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users