On 2023/1/30 22:04, Juan Quintela wrote:
> Jiang Jiacheng <jiangjiach...@huawei.com> wrote:
>> On 2023/1/30 12:28, Juan Quintela wrote:
>>> Jiang Jiacheng <jiangjiach...@huawei.com> wrote:
>>>> To support query migration thread infomation, save and delete
>>>> thread information at thread creation and end.
>>>>
>>>> Signed-off-by: Jiang Jiacheng <jiangjiach...@huawei.com>
>>>
>>> Don't disagree with this, but if we create this on the sending side, why
>>> this is not needed for the multifd_recv_threads?
>>>
>>
>> I only add several threads which i'm interested in into the list till
>> now, whose information will be used for setting cpu affinity for
>> migration thread.
>> For consistency, we can add other threads to the list, but those
>> information won't be used so far.
>
> It is just curiosity, why do you want to set cpu affinity on the source
> but not on destination?
>
Mainly considering the application scenarios. To improve the migration
performance, I want to set cpu affinity for migration thread when
migration a VM whose service threads preempt most of the CPU resources.
So the source side can benefit more from cpu affinity.
And on the destination side, the pressure is lighter (generally),
setting cpu affinity may not be so much useful as source side.
Thanks, Jiang Jiacheng
> Later, Juan.
>
>
