lkml.org 
[lkml]   [2022]   [Mar]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [External] Re: [PATCH] sched/fair: fix broken bandwidth control with nohz_full
From
On 2022/3/28 23:56, Peter Zijlstra wrote:
> On Mon, Mar 28, 2022 at 11:40:25PM +0800, Chengming Zhou wrote:
>
>>> NOHZ_FULL is for use-cases that 'never' intend to go into the kernel,
>>> your use-case actively relies on going into the kernel. Hence the
>>> confusion.
>>
>> In fact, I put a testcase at the end of git message, in which only run
>> a userspace loop workload:
>>
>> cd /sys/fs/cgroup
>> echo "+cpu" > cgroup.subtree_control
>>
>> mkdir test
>> echo "105000 100000" > test/cpu.max
>>
>> echo $$ > test/cgroup.procs
>> taskset -c 1 bash -c "while true; do let i++; done" --> will be throttled
>
> Ofcourse.. I'm arguing that bandiwdth control and NOHZ_FULL are somewhat
> mutually exclusive, use-case wise. So I really don't get why you'd want
> them both.

This problem is found by our VM team, they use bandwidth for overcommit,
share CPUs between two VMs.

>
> NOHZ_FULL says, "I 'never' intend to go to the kernel"

Like VCPU seldom kvm_exit to the kernel, stop tick is helpful for performance,
since kvm_exit is more expensive.

>
> bandwidth control says: "I expect to be sharing the system and must be
> interrupted to not consume too much time", which very much implies: "I
> will go into the kernel".

Yes, agree. If the tasks in the task_group used up quota, they have to
go into the kernel to resched out.

>
> The trade-off we make to make NOHZ_FULL work, makes system enter/exit
> *far* more expensive. There's also people asking to outright kill a task
> that causes entry under NOHZ_FULL.

It's correct that the task under NOHZ_FULL shouldn't often enter/exit.

>
> So yes, you can configure it, but why does it make sense?

I don't know if other people have the same use-case, or is there other
better way to do VMs overcommit and bandwidth...

Thanks.

\
 
 \ /
  Last update: 2022-03-28 18:35    [W:0.104 / U:0.288 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site