Re: Problem with restore/migration with Xen 4.0.0 and Jeremy kernel (2.6.32.12)

Re: Problem with restore/migration with Xen 4.0.0 and Jeremy kernel (2.6.32.12)

[Pasi Kärkkäinen]

Forwarding to xen-devel ..

Jeremy: Have you heard of this before? save/restore/migration
takes 4-10x longer on pvops 2.6.32 compared to xenlinux 2.6.32.

He verified the DEBUG options in kernel .configs are the same.

-- Pasi

On Wed, May 12, 2010 at 05:28:26PM -0400, Pierre POMES wrote:
> Hi,
>
> First sorry for the double posting...
>
> I just built a 2.6.32.10 kernel with Andrew Lyon patches (so it is a  
> "xenlinux" kernel, not a "pvops" kernel).
>
> Live migration and restore operations are between 4 and 10 times faster  
> with this kernel . Furthermore, during live migration, hangs time in  
> domu are shorter (1-2 seconds versus 1 to 15 seconds for a domu with  
> 256M RAM).
>
> Error messages "Error when reading batch size" / "error when buffering  
> batch, finishing" are still in my logs.
>
> Regarding times, all is now similar to what I had with Xen 3.x on top of  
> xenlinux kernels.
>
> Regards,
> Pierre
>
>
>
>> Hi all,
>>
>> I am using Xen 4.0.0 on top of Ubuntu Lucid (amd64), with the Jeremy  
>> kernel taken from git (xen/stable-2.6.32.x branch, 2.6.32.12 when I am  
>> writing this email). This kernel is also used in my domu.
>>
>> I can save a domu without any problem, but restoring it may need from  
>> 2 to 5 minutes, from a 1G checkpoint file (domu has 1GB RAM). There  
>> also errors in /var/log/xen/xend.log, "Error when reading batch size"  
>> and "Error when reading batch size":
>>
>> [2010-05-08 04:23:16 9497] DEBUG (XendDomainInfo:1804) Storing domain  
>> details: {'image/entry': '18446744071587529216', 'console/port': '2',  
>> 'image/loader': 'generic', 'vm':  
>> '/vm/156ea44d-6707-cbe6-2d58-7bea4792dff4',  
>> 'control/platform-feature-multiprocessor-suspend': '1',  
>> 'image/hv-start-low': '18446603336221196288', 'image/guest-os':  
>> 'linux', 'image/virt-base': '18446744071562067968', 'memory/target':  
>> '1048576', 'image/guest-version': '2.6', 'image/pae-mode': 'yes',  
>> 'description': '', 'console/limit': '1048576', 'image/paddr-offset':  
>> '0', 'image/hypercall-page': '18446744071578882048',  
>> 'image/suspend-cancel': '1', 'cpu/0/availability': 'online',  
>> 'image/features/pae-pgdir-above-4gb': '1',  
>> 'image/features/writable-page-tables': '0', 'console/type':  
>> 'xenconsoled', 'name': 'domusample', 'domid': '10',  
>> 'image/xen-version': 'xen-3.0', 'store/port': '1'}
>> [2010-05-08 04:23:16 9497] DEBUG (XendCheckpoint:286)  
>> restore:shadow=0x0, _static_max=0x40000000, _static_min=0x0,
>> [2010-05-08 04:23:16 9497] DEBUG (XendCheckpoint:305) [xc_restore]:  
>> /usr/lib/xen/bin/xc_restore 22 10 1 2 0 0 0 0
>> [2010-05-08 04:23:16 9497] INFO (XendCheckpoint:423) xc_domain_restore  
>> start: p2m_size = 40000
>> [2010-05-08 04:23:16 9497] INFO (XendCheckpoint:423) Reloading memory  
>> pages:   0%
>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) ERROR Internal  
>> error: Error when reading batch size
>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) ERROR Internal  
>> error: error when buffering batch, finishing
>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423)
>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) ^H^H^H^H100%
>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) Memory reloaded  
>> (0 pages)
>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) read VCPU 0
>>
>> Live migration has the same problem, it may need several minutes to  
>> complete. Please note that restore and migration do not fail, but  
>> there are very long.
>>
>> My domu is on top of DRBD, and the config file is:
>>
>>
>> -------------
>> kernel      = '/boot/vmlinuz-2.6.32.12-it-xen'
>> ramdisk     = '/boot/initrd.img-2.6.32.12-it-xen'
>> memory      = '1024'
>>
>> #
>> #  Disk device(s).
>> #
>> root        = '/dev/xvda2 ro'
>> disk        = [
>>                   'drbd:domusampleswap,xvda1,w',
>>                   'drbd:domusampleslash,xvda2,w',
>>               ]
>>
>>
>>
>> #
>> #  Hostname
>> #
>> name        = 'domusample'
>>
>> #
>> #  Networking
>> #
>> vif         = [ 'mac=00:16:3E:58:FC:F9' ]
>>
>> #
>> #  Behaviour
>> #
>> on_poweroff = 'destroy'
>> on_reboot   = 'restart'
>> on_crash    = 'restart'
>>
>> extra = '2 console=hvc0'
>> ----------
>>
>> I do not have any idea here.
>>
>> Did somebody already have (and solve ?)  this issue ?
>>
>> Thanks.
>> Pierre
>>
>>
>> *
>> *
>>
>> _______________________________________________
>> Xen-users mailing list
>> Xen-users@lists.xensource.com
>> http://lists.xensource.com/xen-users
>
>
> _______________________________________________
> Xen-users mailing list
> Xen-users@lists.xensource.com
> http://lists.xensource.com/xen-users

Re: Re: [Xen-users] Problem with restore/migration with Xen 4.0.0 and Jeremy kernel (2.6.32.12)

[Jeremy Fitzhardinge]

On 05/13/2010 06:31 AM, Pasi Kärkkäinen wrote:
> Forwarding to xen-devel ..
>
> Jeremy: Have you heard of this before? save/restore/migration
> takes 4-10x longer on pvops 2.6.32 compared to xenlinux 2.6.32.
>
> He verified the DEBUG options in kernel .configs are the same.
>   

No, I wasn't aware of any big save/restore performance differences.  Is
the difference caused by a pvops dom0 or domU or both?

One materially different thing is that pvops kernels support preemption,
which requires all processes to be frozen before a suspend.  I wonder if
disabling preemption makes a difference (assuming that it is the domU
which is causing the slowdown).

Ah, but the report is that its the restore which is very slow.  Which
suggests that it is the dom0 environment which is causing problems. 
Does "top" show a particular process is very cpu-bound during the
restore?  Or is it IO bound?

    J

> -- Pasi
>
> On Wed, May 12, 2010 at 05:28:26PM -0400, Pierre POMES wrote:
>   
>> Hi,
>>
>> First sorry for the double posting...
>>
>> I just built a 2.6.32.10 kernel with Andrew Lyon patches (so it is a  
>> "xenlinux" kernel, not a "pvops" kernel).
>>
>> Live migration and restore operations are between 4 and 10 times faster  
>> with this kernel . Furthermore, during live migration, hangs time in  
>> domu are shorter (1-2 seconds versus 1 to 15 seconds for a domu with  
>> 256M RAM).
>>
>> Error messages "Error when reading batch size" / "error when buffering  
>> batch, finishing" are still in my logs.
>>
>> Regarding times, all is now similar to what I had with Xen 3.x on top of  
>> xenlinux kernels.
>>
>> Regards,
>> Pierre
>>
>>
>>
>>     
>>> Hi all,
>>>
>>> I am using Xen 4.0.0 on top of Ubuntu Lucid (amd64), with the Jeremy  
>>> kernel taken from git (xen/stable-2.6.32.x branch, 2.6.32.12 when I am  
>>> writing this email). This kernel is also used in my domu.
>>>
>>> I can save a domu without any problem, but restoring it may need from  
>>> 2 to 5 minutes, from a 1G checkpoint file (domu has 1GB RAM). There  
>>> also errors in /var/log/xen/xend.log, "Error when reading batch size"  
>>> and "Error when reading batch size":
>>>
>>> [2010-05-08 04:23:16 9497] DEBUG (XendDomainInfo:1804) Storing domain  
>>> details: {'image/entry': '18446744071587529216', 'console/port': '2',  
>>> 'image/loader': 'generic', 'vm':  
>>> '/vm/156ea44d-6707-cbe6-2d58-7bea4792dff4',  
>>> 'control/platform-feature-multiprocessor-suspend': '1',  
>>> 'image/hv-start-low': '18446603336221196288', 'image/guest-os':  
>>> 'linux', 'image/virt-base': '18446744071562067968', 'memory/target':  
>>> '1048576', 'image/guest-version': '2.6', 'image/pae-mode': 'yes',  
>>> 'description': '', 'console/limit': '1048576', 'image/paddr-offset':  
>>> '0', 'image/hypercall-page': '18446744071578882048',  
>>> 'image/suspend-cancel': '1', 'cpu/0/availability': 'online',  
>>> 'image/features/pae-pgdir-above-4gb': '1',  
>>> 'image/features/writable-page-tables': '0', 'console/type':  
>>> 'xenconsoled', 'name': 'domusample', 'domid': '10',  
>>> 'image/xen-version': 'xen-3.0', 'store/port': '1'}
>>> [2010-05-08 04:23:16 9497] DEBUG (XendCheckpoint:286)  
>>> restore:shadow=0x0, _static_max=0x40000000, _static_min=0x0,
>>> [2010-05-08 04:23:16 9497] DEBUG (XendCheckpoint:305) [xc_restore]:  
>>> /usr/lib/xen/bin/xc_restore 22 10 1 2 0 0 0 0
>>> [2010-05-08 04:23:16 9497] INFO (XendCheckpoint:423) xc_domain_restore  
>>> start: p2m_size = 40000
>>> [2010-05-08 04:23:16 9497] INFO (XendCheckpoint:423) Reloading memory  
>>> pages:   0%
>>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) ERROR Internal  
>>> error: Error when reading batch size
>>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) ERROR Internal  
>>> error: error when buffering batch, finishing
>>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423)
>>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) ^H^H^H^H100%
>>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) Memory reloaded  
>>> (0 pages)
>>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) read VCPU 0
>>>
>>> Live migration has the same problem, it may need several minutes to  
>>> complete. Please note that restore and migration do not fail, but  
>>> there are very long.
>>>
>>> My domu is on top of DRBD, and the config file is:
>>>
>>>
>>> -------------
>>> kernel      = '/boot/vmlinuz-2.6.32.12-it-xen'
>>> ramdisk     = '/boot/initrd.img-2.6.32.12-it-xen'
>>> memory      = '1024'
>>>
>>> #
>>> #  Disk device(s).
>>> #
>>> root        = '/dev/xvda2 ro'
>>> disk        = [
>>>                   'drbd:domusampleswap,xvda1,w',
>>>                   'drbd:domusampleslash,xvda2,w',
>>>               ]
>>>
>>>
>>>
>>> #
>>> #  Hostname
>>> #
>>> name        = 'domusample'
>>>
>>> #
>>> #  Networking
>>> #
>>> vif         = [ 'mac=00:16:3E:58:FC:F9' ]
>>>
>>> #
>>> #  Behaviour
>>> #
>>> on_poweroff = 'destroy'
>>> on_reboot   = 'restart'
>>> on_crash    = 'restart'
>>>
>>> extra = '2 console=hvc0'
>>> ----------
>>>
>>> I do not have any idea here.
>>>
>>> Did somebody already have (and solve ?)  this issue ?
>>>
>>> Thanks.
>>> Pierre
>>>
>>>
>>> *
>>> *
>>>
>>> _______________________________________________
>>> Xen-users mailing list
>>> Xen-users@lists.xensource.com
>>> http://lists.xensource.com/xen-users
>>>       
>>
>> _______________________________________________
>> Xen-users mailing list
>> Xen-users@lists.xensource.com
>> http://lists.xensource.com/xen-users
>>     
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
>
>   

Re: Re: [Xen-users] Problem with restore/migration with Xen 4.0.0 and Jeremy kernel (2.6.32.12)

[Pierre POMES]

Hi Jeremy,
> No, I wasn't aware of any big save/restore performance differences.  Is
> the difference caused by a pvops dom0 or domU or both?
>    

On the domu, I tried a 2.6.32.12 pvops kernel and the standard 
"2.6.32-22-server" from Ubuntu Lucid. It makes no difference.

What is making the difference is using a "xenlinux" kernel in dom0 
(2.6.32.10 with Andrew Lyon patches)

> One materially different thing is that pvops kernels support preemption,
> which requires all processes to be frozen before a suspend.  I wonder if
> disabling preemption makes a difference (assuming that it is the domU
> which is causing the slowdown).
>
> Ah, but the report is that its the restore which is very slow.  Which
> suggests that it is the dom0 environment which is causing problems.
> Does "top" show a particular process is very cpu-bound during the
> restore?  Or is it IO bound?
>    
I tried a restore  a little domu with 256 MB RAM. It tooks around 30s 
with pvops kernel :

root@narbonne:~# time xm restore  pp

real    0m26.905s
user    0m0.070s
sys    0m0.020s

Here is the result of "iostat -c 5 100" during the time of the restore:

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
            0.15    0.00    1.00    0.20    0.15   98.51

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
            0.00    0.00    0.27    0.00    0.14   99.58

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
            0.00    0.00    0.94    0.00    0.35   98.70

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
            0.00    0.00    0.88    0.00    0.46   98.67

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
            0.01    0.00    0.38    0.09    0.15   99.38

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
            0.06    0.00    0.97    0.44    0.30   98.23

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
            0.00    0.00    0.00    0.00    0.08   99.92

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
            0.00    0.00    0.01    0.00    0.00   99.99

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
            0.00    0.00    0.00    0.00    0.00  100.00

So it seems to be neither CPU bound not I/O bound.

I rebooted the server with my xenlinux kernel, and the same restore took :

root@narbonne:~# time xm restore pp

real    0m10.006s
user    0m0.080s
sys    0m0.000s

My domu is on top of DRBD, using the drbd block script in config file.

Live migration can show larger difference between both dom0 kernels : 
10s versus 1 minute for the migration of the same domu.

Hardware: Dell R610 with 32GB RAM, bi quad core, one  RAID10 container 
(hard).

I will compile again my dom0 pvops kernel,  and triple check kernel 
config.  I will also try without DRBD backends.

If I have new results, I will ket you know.

Many thanks,
Pierre


>      J
>
>    
>> -- Pasi
>>
>> On Wed, May 12, 2010 at 05:28:26PM -0400, Pierre POMES wrote:
>>
>>      
>>> Hi,
>>>
>>> First sorry for the double posting...
>>>
>>> I just built a 2.6.32.10 kernel with Andrew Lyon patches (so it is a
>>> "xenlinux" kernel, not a "pvops" kernel).
>>>
>>> Live migration and restore operations are between 4 and 10 times faster
>>> with this kernel . Furthermore, during live migration, hangs time in
>>> domu are shorter (1-2 seconds versus 1 to 15 seconds for a domu with
>>> 256M RAM).
>>>
>>> Error messages "Error when reading batch size" / "error when buffering
>>> batch, finishing" are still in my logs.
>>>
>>> Regarding times, all is now similar to what I had with Xen 3.x on top of
>>> xenlinux kernels.
>>>
>>> Regards,
>>> Pierre
>>>
>>>
>>>
>>>
>>>        
>>>> Hi all,
>>>>
>>>> I am using Xen 4.0.0 on top of Ubuntu Lucid (amd64), with the Jeremy
>>>> kernel taken from git (xen/stable-2.6.32.x branch, 2.6.32.12 when I am
>>>> writing this email). This kernel is also used in my domu.
>>>>
>>>> I can save a domu without any problem, but restoring it may need from
>>>> 2 to 5 minutes, from a 1G checkpoint file (domu has 1GB RAM). There
>>>> also errors in /var/log/xen/xend.log, "Error when reading batch size"
>>>> and "Error when reading batch size":
>>>>
>>>> [2010-05-08 04:23:16 9497] DEBUG (XendDomainInfo:1804) Storing domain
>>>> details: {'image/entry': '18446744071587529216', 'console/port': '2',
>>>> 'image/loader': 'generic', 'vm':
>>>> '/vm/156ea44d-6707-cbe6-2d58-7bea4792dff4',
>>>> 'control/platform-feature-multiprocessor-suspend': '1',
>>>> 'image/hv-start-low': '18446603336221196288', 'image/guest-os':
>>>> 'linux', 'image/virt-base': '18446744071562067968', 'memory/target':
>>>> '1048576', 'image/guest-version': '2.6', 'image/pae-mode': 'yes',
>>>> 'description': '', 'console/limit': '1048576', 'image/paddr-offset':
>>>> '0', 'image/hypercall-page': '18446744071578882048',
>>>> 'image/suspend-cancel': '1', 'cpu/0/availability': 'online',
>>>> 'image/features/pae-pgdir-above-4gb': '1',
>>>> 'image/features/writable-page-tables': '0', 'console/type':
>>>> 'xenconsoled', 'name': 'domusample', 'domid': '10',
>>>> 'image/xen-version': 'xen-3.0', 'store/port': '1'}
>>>> [2010-05-08 04:23:16 9497] DEBUG (XendCheckpoint:286)
>>>> restore:shadow=0x0, _static_max=0x40000000, _static_min=0x0,
>>>> [2010-05-08 04:23:16 9497] DEBUG (XendCheckpoint:305) [xc_restore]:
>>>> /usr/lib/xen/bin/xc_restore 22 10 1 2 0 0 0 0
>>>> [2010-05-08 04:23:16 9497] INFO (XendCheckpoint:423) xc_domain_restore
>>>> start: p2m_size = 40000
>>>> [2010-05-08 04:23:16 9497] INFO (XendCheckpoint:423) Reloading memory
>>>> pages:   0%
>>>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) ERROR Internal
>>>> error: Error when reading batch size
>>>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) ERROR Internal
>>>> error: error when buffering batch, finishing
>>>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423)
>>>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) ^H^H^H^H100%
>>>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) Memory reloaded
>>>> (0 pages)
>>>> [2010-05-08 04:25:53 9497] INFO (XendCheckpoint:423) read VCPU 0
>>>>
>>>> Live migration has the same problem, it may need several minutes to
>>>> complete. Please note that restore and migration do not fail, but
>>>> there are very long.
>>>>
>>>> My domu is on top of DRBD, and the config file is:
>>>>
>>>>
>>>> -------------
>>>> kernel      = '/boot/vmlinuz-2.6.32.12-it-xen'
>>>> ramdisk     = '/boot/initrd.img-2.6.32.12-it-xen'
>>>> memory      = '1024'
>>>>
>>>> #
>>>> #  Disk device(s).
>>>> #
>>>> root        = '/dev/xvda2 ro'
>>>> disk        = [
>>>>                    'drbd:domusampleswap,xvda1,w',
>>>>                    'drbd:domusampleslash,xvda2,w',
>>>>                ]
>>>>
>>>>
>>>>
>>>> #
>>>> #  Hostname
>>>> #
>>>> name        = 'domusample'
>>>>
>>>> #
>>>> #  Networking
>>>> #
>>>> vif         = [ 'mac=00:16:3E:58:FC:F9' ]
>>>>
>>>> #
>>>> #  Behaviour
>>>> #
>>>> on_poweroff = 'destroy'
>>>> on_reboot   = 'restart'
>>>> on_crash    = 'restart'
>>>>
>>>> extra = '2 console=hvc0'
>>>> ----------
>>>>
>>>> I do not have any idea here.
>>>>
>>>> Did somebody already have (and solve ?)  this issue ?
>>>>
>>>> Thanks.
>>>> Pierre
>>>>
>>>>
>>>> *
>>>> *
>>>>
>>>> _______________________________________________
>>>> Xen-users mailing list
>>>> Xen-users@lists.xensource.com
>>>> http://lists.xensource.com/xen-users
>>>>
>>>>          
>>> _______________________________________________
>>> Xen-users mailing list
>>> Xen-users@lists.xensource.com
>>> http://lists.xensource.com/xen-users
>>>
>>>        
>> _______________________________________________
>> Xen-devel mailing list
>> Xen-devel@lists.xensource.com
>> http://lists.xensource.com/xen-devel
>>
>>
>>      
>