[PVE-User] linux idle cpu overhead in kvm - old issue, but still there in 2020...

[PVE-User] linux idle cpu overhead in kvm - old issue, but still there in 2020...

[Roland]

hello,

i have found that there is an old bug still around in linux, which is
causing quite an amount of unnecessary cpu consumption in kvm/proxmox,
and thus, wasting precious power.

i run some proxmox installations on older systems and on those, it's
quite significant difference.

on the slowest system, a single debian 10 VM , kvm process is at 20% cpu
(VM is 100% idle) when this issue is present.

if i change VMs machine type from i440fx(default) to q35 the problem
goes away.

the same applies when running "powertop --auto-tune" inside the guest
(with i440fx type - enable autosuspend for usb-controller + tablet device).

on some L5630 machine, in proxmox summary i see "CPU usage" drop from
10% to <1%.

see:
https://bugzilla.redhat.com/show_bug.cgi?id=478317
https://bugzilla.redhat.com/show_bug.cgi?id=949547

i guess this information could make a difference for people who run a
large amount of virtual machines or use older systems/cpu's.

on most recent cpu's, i think the difference is not that big.

anyway, i really wonder how linux bugs have such great survival
capability....

regards
roland

more references:
https://lists.gnu.org/archive/html/qemu-devel/2010-04/msg00149.html
https://www.redhat.com/archives/vfio-users/2015-November/msg00159.html

Re: [PVE-User] linux idle cpu overhead in kvm - old issue, but still there in 2020...

[Atila Vasconcelos]

Wow, I just tried this at my servers (very old Dell PowerEdge 2950);

The results are impressive!

8o


ABV


On 2020-07-13 3:15 a.m., Roland wrote:
> hello,
>
> i have found that there is an old bug still around in linux, which is
> causing quite an amount of unnecessary cpu consumption in kvm/proxmox,
> and thus, wasting precious power.
>
> i run some proxmox installations on older systems and on those, it's
> quite significant difference.
>
> on the slowest system, a single debian 10 VM , kvm process is at 20% cpu
> (VM is 100% idle) when this issue is present.
>
> if i change VMs machine type from i440fx(default) to q35 the problem
> goes away.
>
> the same applies when running "powertop --auto-tune" inside the guest
> (with i440fx type - enable autosuspend for usb-controller + tablet 
> device).
>
> on some L5630 machine, in proxmox summary i see "CPU usage" drop from
> 10% to <1%.
>
> see:
> https://bugzilla.redhat.com/show_bug.cgi?id=478317
> https://bugzilla.redhat.com/show_bug.cgi?id=949547
>
> i guess this information could make a difference for people who run a
> large amount of virtual machines or use older systems/cpu's.
>
> on most recent cpu's, i think the difference is not that big.
>
> anyway, i really wonder how linux bugs have such great survival
> capability....
>
> regards
> roland
>
> more references:
> https://lists.gnu.org/archive/html/qemu-devel/2010-04/msg00149.html
> https://www.redhat.com/archives/vfio-users/2015-November/msg00159.html
>
>
> _______________________________________________
> pve-user mailing list
> pve-user@lists.proxmox.com
> https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-user
>

Re: [PVE-User] linux idle cpu overhead in kvm - old issue, but still there in 2020...

[Roland]

nice :)

what guest OS do you use and which showed the problem ?

as i think that q35 will not be default in kvm or proxmox anytime soon,
shouldn't we perhaps file a bug report for every distro affected ?

is there someone who likes to work with this (help testing,
writing/tracking bug reports...) ?

could perhaps save some tons of CO2 ....

roland


Am 14.07.20 um 18:05 schrieb Atila Vasconcelos:
> Wow, I just tried this at my servers (very old Dell PowerEdge 2950);
>
> The results are impressive!
>
> 8o
>
>
> ABV
>
>
> On 2020-07-13 3:15 a.m., Roland wrote:
>> hello,
>>
>> i have found that there is an old bug still around in linux, which is
>> causing quite an amount of unnecessary cpu consumption in kvm/proxmox,
>> and thus, wasting precious power.
>>
>> i run some proxmox installations on older systems and on those, it's
>> quite significant difference.
>>
>> on the slowest system, a single debian 10 VM , kvm process is at 20% cpu
>> (VM is 100% idle) when this issue is present.
>>
>> if i change VMs machine type from i440fx(default) to q35 the problem
>> goes away.
>>
>> the same applies when running "powertop --auto-tune" inside the guest
>> (with i440fx type - enable autosuspend for usb-controller + tablet
>> device).
>>
>> on some L5630 machine, in proxmox summary i see "CPU usage" drop from
>> 10% to <1%.
>>
>> see:
>> https://bugzilla.redhat.com/show_bug.cgi?id=478317
>> https://bugzilla.redhat.com/show_bug.cgi?id=949547
>>
>> i guess this information could make a difference for people who run a
>> large amount of virtual machines or use older systems/cpu's.
>>
>> on most recent cpu's, i think the difference is not that big.
>>
>> anyway, i really wonder how linux bugs have such great survival
>> capability....
>>
>> regards
>> roland
>>
>> more references:
>> https://lists.gnu.org/archive/html/qemu-devel/2010-04/msg00149.html
>> https://www.redhat.com/archives/vfio-users/2015-November/msg00159.html
>>
>>
>> _______________________________________________
>> pve-user mailing list
>> pve-user@lists.proxmox.com
>> https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-user
>>
>
> _______________________________________________
> pve-user mailing list
> pve-user@lists.proxmox.com
> https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-user
>

Re: [PVE-User] linux idle cpu overhead in kvm - old issue, but still there in 2020...

[Roland]

apparently, this issue will get fixed upstream (in systemd package):

https://github.com/systemd/systemd/pull/353#issuecomment-658810289
https://github.com/systemd/systemd/pull/16476

roland

Am 15.07.20 um 19:18 schrieb Roland:
>
> >if i change VMs machine type from i440fx(default) to q35 the problem
> >goes away.
> >
> >the same applies when running "powertop --auto-tune" inside the guest
> >(with i440fx type - enable autosuspend for usb-controller + tablet
> device).
>
> third option (for VMs without gui):
>
> set "use tablet for pointer " to "No" in VM's options
>
> https://forum.proxmox.com/threads/high-cpu-usage-of-usb-tablet-device-in-debian.45307/
>
> https://forum.proxmox.com/threads/use-tablet-for-pointer-option-causing-cpu-usage-on-linux.54084/
>
> roland
>
>
> Am 14.07.20 um 19:16 schrieb Roland:
>> nice :)
>>
>> what guest OS do you use and which showed the problem ?
>>
>> as i think that q35 will not be default in kvm or proxmox anytime
>> soon, shouldn't we perhaps file a bug report for every distro affected ?
>>
>> is there someone who likes to work with this (help testing,
>> writing/tracking bug reports...) ?
>>
>> could perhaps save some tons of CO2 ....
>>
>> roland
>>
>>
>> Am 14.07.20 um 18:05 schrieb Atila Vasconcelos:
>>> Wow, I just tried this at my servers (very old Dell PowerEdge 2950);
>>>
>>> The results are impressive!
>>>
>>> 8o
>>>
>>>
>>> ABV
>>>
>>>
>>> On 2020-07-13 3:15 a.m., Roland wrote:
>>>> hello,
>>>>
>>>> i have found that there is an old bug still around in linux, which is
>>>> causing quite an amount of unnecessary cpu consumption in kvm/proxmox,
>>>> and thus, wasting precious power.
>>>>
>>>> i run some proxmox installations on older systems and on those, it's
>>>> quite significant difference.
>>>>
>>>> on the slowest system, a single debian 10 VM , kvm process is at
>>>> 20% cpu
>>>> (VM is 100% idle) when this issue is present.
>>>>
>>>> if i change VMs machine type from i440fx(default) to q35 the problem
>>>> goes away.
>>>>
>>>> the same applies when running "powertop --auto-tune" inside the guest
>>>> (with i440fx type - enable autosuspend for usb-controller + tablet
>>>> device).
>>>>
>>>> on some L5630 machine, in proxmox summary i see "CPU usage" drop from
>>>> 10% to <1%.
>>>>
>>>> see:
>>>> https://bugzilla.redhat.com/show_bug.cgi?id=478317
>>>> https://bugzilla.redhat.com/show_bug.cgi?id=949547
>>>>
>>>> i guess this information could make a difference for people who run a
>>>> large amount of virtual machines or use older systems/cpu's.
>>>>
>>>> on most recent cpu's, i think the difference is not that big.
>>>>
>>>> anyway, i really wonder how linux bugs have such great survival
>>>> capability....
>>>>
>>>> regards
>>>> roland
>>>>
>>>> more references:
>>>> https://lists.gnu.org/archive/html/qemu-devel/2010-04/msg00149.html
>>>> https://www.redhat.com/archives/vfio-users/2015-November/msg00159.html
>>>>
>>>>
>>>> _______________________________________________
>>>> pve-user mailing list
>>>> pve-user@lists.proxmox.com
>>>> https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-user
>>>>
>>>
>>> _______________________________________________
>>> pve-user mailing list
>>> pve-user@lists.proxmox.com
>>> https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-user
>>>