Re: [PATCH 1/2] timer: Avoid waking up an idle-core by migrate running timer

From: Eric Dumazet <eric.dumazet@gmail.com>
To: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>,
	viresh kumar <viresh.kumar@linaro.org>,
	Ingo Molnar <mingo@redhat.com>,
	linaro-kernel@lists.linaro.org,
	LKML <linux-kernel@vger.kernel.org>,
	Steven Miao <realmz6@gmail.com>,
	shashim@codeaurora.org, Frederic Weisbecker <fweisbec@gmail.com>,
	"Paul E. McKenney" <paulmck@linux.vnet.ibm.com>,
	cl@kernel.org
Subject: Re: [PATCH 1/2] timer: Avoid waking up an idle-core by migrate running timer
Date: Sat, 25 Apr 2015 11:37:30 -0700	[thread overview]
Message-ID: <1429987050.22254.187.camel@edumazet-glaptop2.roam.corp.google.com> (raw)
In-Reply-To: <alpine.DEB.2.11.1504231439300.13914@nanos>

On Thu, 2015-04-23 at 14:45 +0200, Thomas Gleixner wrote:

> You definitely have a point from the high throughput networking
> perspective.
> 
> Though in a power optimizing scenario with minimal network traffic
> this might be the wrong decision. We have to gather data from the
> power maniacs whether this matters or not. The FULL_NO_HZ camp might
> be pretty unhappy about the above.

Sure, I understand.

To make this clear, here the profile on a moderately loaded TCP server,
pushing ~20Gbits of data. Most of TCP output is ACK clock driven (thus
from softirq context).

(using regular sendmsg() system calls, that why the
get_nohz_timer_target() is 'only' second in the profile, but add the
find_next_bit() to it and this is very close being at first position)

   PerfTop:    4712 irqs/sec  kernel:96.7%  exact:  0.0% [4000Hz cycles],  (all, 72 CPUs)
-----------------------------------------------------------------------------------------
    10.16%  [kernel]          [k] copy_user_enhanced_fast_string            
     5.66%  [kernel]          [k] get_nohz_timer_target                     
     5.59%  [kernel]          [k] _raw_spin_lock                            
     2.53%  [kernel]          [k] __netif_receive_skb_core                  
     2.27%  [kernel]          [k] find_next_bit                             
     1.90%  [kernel]          [k] tcp_ack                                   

Maybe a reasonable heuristic would be to
change /proc/sys/kernel/timer_migration default to 0 on hosts with more
than 32 cpus.

profile with timer_migration = 0

   PerfTop:    3656 irqs/sec  kernel:94.3%  exact:  0.0% [4000Hz cycles],  (all, 72 CPUs)
-----------------------------------------------------------------------------------------
    13.95%  [kernel]          [k] copy_user_enhanced_fast_string            
     4.65%  [kernel]          [k] _raw_spin_lock                            
     2.57%  [kernel]          [k] __netif_receive_skb_core                  
     2.33%  [kernel]          [k] tcp_ack               

Thanks.