Messages in this thread | | | From | Wanpeng Li <> | Date | Wed, 23 Oct 2019 16:16:55 +0800 | Subject | Re: [PATCH v2] sched/nohz: Optimize get_nohz_timer_target() |
| |
On Fri, 28 Jun 2019 at 09:10, Frederic Weisbecker <frederic@kernel.org> wrote: > > On Fri, Jun 28, 2019 at 08:43:12AM +0800, Wanpeng Li wrote: > > From: Wanpeng Li <wanpengli@tencent.com> > > > > On a machine, cpu 0 is used for housekeeping, the other 39 cpus in the > > same socket are in nohz_full mode. We can observe huge time burn in the > > loop for seaching nearest busy housekeeper cpu by ftrace. > > > > 2) | get_nohz_timer_target() { > > 2) 0.240 us | housekeeping_test_cpu(); > > 2) 0.458 us | housekeeping_test_cpu(); > > > > ... > > > > 2) 0.292 us | housekeeping_test_cpu(); > > 2) 0.240 us | housekeeping_test_cpu(); > > 2) 0.227 us | housekeeping_any_cpu(); > > 2) + 43.460 us | } > > > > This patch optimizes the searching logic by finding a nearest housekeeper > > cpu in the housekeeping cpumask, it can minimize the worst searching time > > from ~44us to < 10us in my testing. In addition, the last iterated busy > > housekeeper can become a random candidate while current CPU is a better > > fallback if it is a housekeeper. > > > > Cc: Ingo Molnar <mingo@redhat.com> > > Cc: Peter Zijlstra <peterz@infradead.org> > > Cc: Frederic Weisbecker <frederic@kernel.org> > > Cc: Thomas Gleixner <tglx@linutronix.de> > > Signed-off-by: Wanpeng Li <wanpengli@tencent.com> > > Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Hi Thomas,
I didn't see your refactor to get_nohz_timer_target() which you mentioned in IRC after four months, I can observe cyclictest drop from 4~5us to 8us in kvm guest(we offload the lapic timer emulation to housekeeping cpu to avoid timer fire external interrupt on the pCPU which vCPU resident incur a vCPU vmexit) w/o this patch in the case of there is no busy housekeeping cpu. The score can be recovered after I give stress to create a busy housekeeping cpu.
Could you consider applying this patch for temporary since I'm not sure when the refactor can be ready.
Wanpeng
| |