Linux Virtualization - Apr 2015 - [PATCH v15 16/16] unfair qspinlock: a queue based unfair lock

On 04/09/2015 03:01 AM, Peter Zijlstra wrote:
> On Wed, Apr 08, 2015 at 02:32:19PM -0400, Waiman Long wrote:
>> For a virtual guest with the qspinlock patch, a simple unfair byte lock
>> will be used if PV spinlock is not configured in or the hypervisor
>> isn't either KVM or Xen. The byte lock works fine with small guest
>> of just a few vCPUs. On a much larger guest, however, byte lock can
>> have serious performance problem.
> 
> Who cares?
There are some people out there running guests with dozens
of vCPUs. If the code exists to make those setups run better,
is there a good reason not to use it?

Having said that, only KVM and Xen seem to support very
large guests, and PV spinlock is available there.

I believe both VMware and Hyperv have a 32 VCPU limit, anyway.

-- 
All rights reversed

On Thu, Apr 09, 2015 at 09:16:24AM -0400, Rik van Riel
wrote:
> On 04/09/2015 03:01 AM, Peter Zijlstra wrote:
> > On Wed, Apr 08, 2015 at 02:32:19PM -0400, Waiman Long wrote:
> >> For a virtual guest with the qspinlock patch, a simple unfair byte
lock
> >> will be used if PV spinlock is not configured in or the hypervisor
> >> isn't either KVM or Xen. The byte lock works fine with small
guest
> >> of just a few vCPUs. On a much larger guest, however, byte lock
can
> >> have serious performance problem.
> > 
> > Who cares?
> 
> There are some people out there running guests with dozens
> of vCPUs. If the code exists to make those setups run better,
> is there a good reason not to use it?
Well use paravirt, !paravirt stuff sucks performance wise anyhow.

The question really is: is the added complexity worth the maintenance
burden. And I'm just not convinced !paravirt virt is a performance
critical target.
> Having said that, only KVM and Xen seem to support very
> large guests, and PV spinlock is available there.
> 
> I believe both VMware and Hyperv have a 32 VCPU limit, anyway.
Don't we have Hyperv paravirt drivers? They could add support for
paravirt spinlocks too.

On 04/09/2015 10:13 AM, Peter Zijlstra wrote:
> On Thu, Apr 09, 2015 at 09:16:24AM -0400, Rik van Riel wrote:
>> On 04/09/2015 03:01 AM, Peter Zijlstra wrote:
>>> On Wed, Apr 08, 2015 at 02:32:19PM -0400, Waiman Long wrote:
>>>> For a virtual guest with the qspinlock patch, a simple unfair
byte lock
>>>> will be used if PV spinlock is not configured in or the
hypervisor
>>>> isn't either KVM or Xen. The byte lock works fine with
small guest
>>>> of just a few vCPUs. On a much larger guest, however, byte lock
can
>>>> have serious performance problem.
>>>
>>> Who cares?
>>
>> There are some people out there running guests with dozens
>> of vCPUs. If the code exists to make those setups run better,
>> is there a good reason not to use it?
> 
> Well use paravirt, !paravirt stuff sucks performance wise anyhow.
> 
> The question really is: is the added complexity worth the maintenance
> burden. And I'm just not convinced !paravirt virt is a performance
> critical target.
Fair enough.

On 04/09/2015 10:13 AM, Peter Zijlstra wrote:
> On Thu, Apr 09, 2015 at 09:16:24AM -0400, Rik van Riel wrote:
>> On 04/09/2015 03:01 AM, Peter Zijlstra wrote:
>>> On Wed, Apr 08, 2015 at 02:32:19PM -0400, Waiman Long wrote:
>>>> For a virtual guest with the qspinlock patch, a simple unfair
byte lock
>>>> will be used if PV spinlock is not configured in or the
hypervisor
>>>> isn't either KVM or Xen. The byte lock works fine with
small guest
>>>> of just a few vCPUs. On a much larger guest, however, byte lock
can
>>>> have serious performance problem.
>>> Who cares?
>> There are some people out there running guests with dozens
>> of vCPUs. If the code exists to make those setups run better,
>> is there a good reason not to use it?
> Well use paravirt, !paravirt stuff sucks performance wise anyhow.
>
> The question really is: is the added complexity worth the maintenance
> burden. And I'm just not convinced !paravirt virt is a performance
> critical target.
I am just thinking that the unfair qspinlock is better performing than 
the simple byte lock. However, my current priority is to get native and 
PV qspinlock upstream. The unfair qspinlock can certainly wait.

Cheers,
Longman