Xen devel - Nov 2006 - xen and latency - sending faxes via capi (fritzcard USB in domU)

Hi,

I am having the following problem: I am sending faxes via HylaFax over a
FritzCard USB in a domU. The Card is attached to an USB controller, which is
passed to the domU.

I had the bvt scheduler, and had no Problems with it. I switched to the credit
scheduler a few weeks ago. Now I have the problem, that sometimes faxes
don''t get transmitted complete.

The box has a very low load - none of the domains has much to do, and it is a
dual Opteron. Sending faxes is time critical - AVM implements the capi in the
kernel space, to have lower latency.

I know, that virtualization is very bad for latency critical tasks.

I tried to set the weigh of the fax sending domain to a very high level (4000)
and the other domains to 128, but it didn''t help.

How can I grant a given latency to a domU? Or yould it be best, to put the capi
stuff in dom0? I don''t like to put this into dom0, because rebooting
dom0 is much more difficult, than domU.

Johnny

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

> I am having the following problem: I am sending faxes via HylaFax over
a
> FritzCard USB in a domU. The Card is attached to an USB controller,
which
> is passed to the domU.
> 
> I had the bvt scheduler, and had no Problems with it. I switched to
the
> credit scheduler a few weeks ago. Now I have the problem, that
sometimes
> faxes don''t get transmitted complete.
> 
> The box has a very low load - none of the domains has much to do, and
it is
> a dual Opteron. Sending faxes is time critical - AVM implements the
capi in
> the kernel space, to have lower latency.
> 
> I know, that virtualization is very bad for latency critical tasks.
> 
> I tried to set the weigh of the fax sending domain to a very high
level
> (4000) and the other domains to 128, but it didn''t help.
> 
> How can I grant a given latency to a domU? Or yould it be best, to put
the
> capi stuff in dom0? I don''t like to put this into dom0, because
rebooting
> dom0 is much more difficult, than domU.
What version of xen are you running?
It might be worth confirming that it''s scheduling issues causing the
problem by arranging to pin your domains'' VCPUs such that the fax
domain
is guaranteed a physical CPU to itself. 

It also might be worth trying latest xen-unstable as there have been a
few minor fixes to the credit scheduler since 3.0.3-0 that might effect
this scenario.

Ian 




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

Hi,

thank you for your quick reply.

Am Samstag, 4. November 2006 18:13 schrieb Ian Pratt:
> [...]
> 
> What version of xen are you running?
Oh sorry, I forgot.

xen_changeset          : Thu Oct 12 15:28:59 2006 +0100 11748:af1aa35265eb

kernel 2.6.16.29
> It might be worth confirming that it''s scheduling issues causing
the
> problem by arranging to pin your domains'' VCPUs such that the fax
domain
> is guaranteed a physical CPU to itself. 
> 
> It also might be worth trying latest xen-unstable as there have been a
> few minor fixes to the credit scheduler since 3.0.3-0 that might effect
> this scenario.
I''ll try that. Is the current xen unstable safe for production systems?
I''ve been using unstable, and had no problems. Have there been any
dangerous patches?

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

This is odd. If there isn''t much load on the box, then it
shouldn''t matter very much which scheduler is being used.

Since BVT was removed from unstable some time ago, I assume
that your transition from BVT to Credit involved other changes
or perhaps that you''re running an older version of the credit
scheduler. As Ian pointed out, there have been a few changes
in the credit scheduler recently and many since BVT was
removed which may be relevant here.

What version of BVT and credit are you running?


On Fri, Nov 03, 2006 at 07:07:07PM +0100, Hans-Christian Armingeon
wrote:
> Hi,
> 
> I am having the following problem: I am sending faxes via HylaFax over a
FritzCard USB in a domU. The Card is attached to an USB controller, which is
passed to the domU.
> 
> I had the bvt scheduler, and had no Problems with it. I switched to the
credit scheduler a few weeks ago. Now I have the problem, that sometimes faxes
don''t get transmitted complete.
> 
> The box has a very low load - none of the domains has much to do, and it is
a dual Opteron. Sending faxes is time critical - AVM implements the capi in the
kernel space, to have lower latency.
> 
> I know, that virtualization is very bad for latency critical tasks.
> 
> I tried to set the weigh of the fax sending domain to a very high level
(4000) and the other domains to 128, but it didn''t help.
> 
> How can I grant a given latency to a domU? Or yould it be best, to put the
capi stuff in dom0? I don''t like to put this into dom0, because
rebooting dom0 is much more difficult, than domU.
> 
> Johnny
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Am Sonntag, 5. November 2006 17:33 schrieb Ian Pratt:
> > I tried to upgrade xen, but it doesn''t accept the PCI device
option,
> as I''m
> > writing in "[Xen-users] pciback: Error: too many values to
unpack". Do
> you
> > have any ideas, what changed?
> > 
> > I would like to update the xen system, but the pci passthrough is
> cruical.
> 
> Which cset are you using? What version were you using before?
xen_changeset          : Thu Oct 12 15:28:59 2006 +0100 11748:af1aa35265eb is
working with this config. the current doesn''t.

#usb
pci = [ ''0000:03:00.0'', ''0000:03:00.1'' ]

Full config attached.
> What syntax are you using in the config file? Please post the error log
> from /var/log/xen*
xend.log:

Traceback (most recent call last):
  File
"/data/src/xen/build/xen-unstable.hg-3.0-20061104.2/debian/xen-unstable.hg-3.0/usr/lib/python/xen/util/xmlrpclib2.py",
line 179, in _marshaled_dispatch
  File "/usr/lib/python2.4/SimpleXMLRPCServer.py", line 406, in
_dispatch
    return func(*params)
  File
"/data/src/xen/build/xen-unstable.hg-3.0-20061104.2/debian/xen-unstable.hg-3.0/usr/lib/python/xen/xend/server/XMLRPCServer.py",
line 64, in domain_create
  File
"/data/src/xen/build/xen-unstable.hg-3.0-20061104.2/debian/xen-unstable.hg-3.0/usr/lib/python/xen/xend/XendDomain.py",
line 768, in domain_create
  File
"/data/src/xen/build/xen-unstable.hg-3.0-20061104.2/debian/xen-unstable.hg-3.0/usr/lib/python/xen/xend/XendDomainInfo.py",
line 147, in create
  File
"/data/src/xen/build/xen-unstable.hg-3.0-20061104.2/debian/xen-unstable.hg-3.0/usr/lib/python/xen/xend/XendConfig.py",
line 285, in __init__
  File
"/data/src/xen/build/xen-unstable.hg-3.0-20061104.2/debian/xen-unstable.hg-3.0/usr/lib/python/xen/xend/XendConfig.py",
line 455, in _populate_from_sxp
ValueError: too many values to unpack
[2006-11-05 17:44:57 xend.XendDomainInfo 3181] DEBUG (XendDomainInfo:146)
XendDomainInfo.create([''vm'', [''name'',
''themisto''], [''memory'', 256],
[''on_poweroff'', ''destroy''],
[''on_reboot'', ''restart''],
[''on_crash'', ''restart''],
[''vcpus'', 1], [''on_xend_start'',
''ignore''], [''on_xend_stop'',
''ignore''], [''image'',
[''linux'', [''kernel'',
''/boot/vmlinuz-nosmp-xenU''], [''ramdisk'',
''/boot/initrd.img-nosmp-xenU''], [''ip'',
''192.168.5.59:1.2.3.4:192.168.5.15:255.255.255.0::eth0:off''],
[''root'', ''/dev/sda1 ro'']]],
[''device'', [''vbd'',
[''uname'',
''phy:promotheus500g/themisto-xenU-root''],
[''dev'', ''sda1''], [''mode'',
''w'']]], [''device'', [''vbd'',
[''uname'',
''phy:promotheus500g/themisto-xenU-swap''],
[''dev'', ''sda2''], [''mode'',
''w'']]], [''device'', [''vbd'',
[''uname'', ''phy:promotheus500g/modules''],
[''dev'', ''hda1''], [''mode'',
''r'']]], [''device'', [''pci'',
[''dev'', [''domain'', ''0x0''],
[''bus'', ''0x03''], [''slot'',
''0x00''], [''func'',
''0x0'']]]], [''device'',
[''vif'', [''bridge'',
''xenbr0''], [''mac'',
''00:0c:8d:00:0a:10'']]]])
[2006-11-05 17:44:57 xend 3181] DEBUG (XendConfig:457) XendConfig: reading
device: {''uname'':
''phy:promotheus500g/themisto-xenU-root'',
''mode'': ''w'', ''dev'':
''sda1''}
[2006-11-05 17:44:57 xend 3181] DEBUG (XendConfig:457) XendConfig: reading
device: {''uname'':
''phy:promotheus500g/themisto-xenU-swap'',
''mode'': ''w'', ''dev'':
''sda2''}
[2006-11-05 17:44:57 xend 3181] DEBUG (XendConfig:457) XendConfig: reading
device: {''uname'':
''phy:promotheus500g/modules'', ''mode'':
''r'', ''dev'': ''hda1''}
[2006-11-05 17:44:57 xend 3181] ERROR (xmlrpclib2:200) too many values to unpack
Traceback (most recent call last):
  File
"/data/src/xen/build/xen-unstable.hg-3.0-20061104.2/debian/xen-unstable.hg-3.0/usr/lib/python/xen/util/xmlrpclib2.py",
line 179, in _marshaled_dispatch
  File "/usr/lib/python2.4/SimpleXMLRPCServer.py", line 406, in
_dispatch
    return func(*params)
  File
"/data/src/xen/build/xen-unstable.hg-3.0-20061104.2/debian/xen-unstable.hg-3.0/usr/lib/python/xen/xend/server/XMLRPCServer.py",
line 64, in domain_create
  File
"/data/src/xen/build/xen-unstable.hg-3.0-20061104.2/debian/xen-unstable.hg-3.0/usr/lib/python/xen/xend/XendDomain.py",
line 768, in domain_create
  File
"/data/src/xen/build/xen-unstable.hg-3.0-20061104.2/debian/xen-unstable.hg-3.0/usr/lib/python/xen/xend/XendDomainInfo.py",
line 147, in create
  File
"/data/src/xen/build/xen-unstable.hg-3.0-20061104.2/debian/xen-unstable.hg-3.0/usr/lib/python/xen/xend/XendConfig.py",
line 285, in __init__
  File
"/data/src/xen/build/xen-unstable.hg-3.0-20061104.2/debian/xen-unstable.hg-3.0/usr/lib/python/xen/xend/XendConfig.py",
line 455, in _populate_from_sxp
ValueError: too many values to unpack


xen-hotplug.log:

Nothing to flush.


Well, I''m using a script, that adjusts the weighs of all domains. I had
the fax domain at 4096, and the next impoortant system at 256. I set the
priority now to 65000 of the fax domain.
> 
> Ian 
> 
> 
Thank you for your help, in sorting out this issue,

Johnny


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

Am Sonntag, 5. November 2006 17:30 schrieb Emmanuel
Ackaouy:
> This is odd. If there isn''t much load on the box, then it
> shouldn''t matter very much which scheduler is being used.
> 
> Since BVT was removed from unstable some time ago, I assume
> that your transition from BVT to Credit involved other changes
> or perhaps that you''re running an older version of the credit
> scheduler. As Ian pointed out, there have been a few changes
> in the credit scheduler recently and many since BVT was
> removed which may be relevant here.
> 
> What version of BVT and credit are you running?
On 21. October I switched from bvt to credit, because I didn''t spend
much time in tuning the sedf scheduler. I''ve never used the sedf
scheduler in production.
I think that was the bvt scheduler from xen unstable from 20061011.

I tried to use current xen unstable, but PCI passthrough doesn''t work
with my config.

Johnny


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

On Sun, Nov 05, 2006 at 05:53:19PM +0100, Hans-Christian Armingeon
wrote:
> > What version of BVT and credit are you running?
> 
> On 21. October I switched from bvt to credit, because I didn''t
spend much time in tuning the sedf scheduler. I''ve never used the sedf
scheduler in production.
> I think that was the bvt scheduler from xen unstable from 20061011.
I don''t get it.

BVT was removed from xen-unstable in late August:

changeset:   11241:26b673aeff8b
user:        kfraser@localhost.localdomain
date:        Wed Aug 23 17:14:38 2006 +0100
summary:     Remove all traces of the obsolete BVT scheduler.

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

Am Sonntag, 5. November 2006 18:09 schrieb Emmanuel
Ackaouy:
> On Sun, Nov 05, 2006 at 05:53:19PM +0100, Hans-Christian Armingeon wrote:
> > > What version of BVT and credit are you running?
> > 
> > On 21. October I switched from bvt to credit, because I
didn''t spend much time in tuning the sedf scheduler. I''ve
never used the sedf scheduler in production.
> > I think that was the bvt scheduler from xen unstable from 20061011.
> 
> I don''t get it.
> 
> BVT was removed from xen-unstable in late August:
Well, I changed the removed the sched=bvt about this time.

Johnny
> 
> changeset:   11241:26b673aeff8b
> user:        kfraser@localhost.localdomain
> date:        Wed Aug 23 17:14:38 2006 +0100
> summary:     Remove all traces of the obsolete BVT scheduler.
> 
> 
_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

On 4 Nov 2006 at 22:43, Hans-Christian Armingeon wrote:
> I''ll try that. Is the current xen unstable safe for production
systems? I''ve been using unstable, and had no problems. Have there been
any dangerous patches?
> 
I think nobody ever said, scheduled DomUs are suitable for real-time
applications.
Some USB drivers however assume they have full control over the CPU. Probably
not
when they are running in a DomU kernel...

Regards,
Ulrich


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