Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220712082036.5130-1-wuyun.abel@bytedance.com> <20220712082036.5130-2-wuyun.abel@bytedance.com>
In-Reply-To: <20220712082036.5130-2-wuyun.abel@bytedance.com>
From:   Chen Yu <yu.chen.surf@gmail.com>
Date:   Wed, 10 Aug 2022 21:50:06 +0800
Message-ID: <CADjb_WRdvKm=uJGe2zE+D6fHkuCX4i+WkrWqz1A3VXmtdKN_Tg@mail.gmail.com>
Subject: Re: [PATCH 1/5] sched/fair: ignore SIS_UTIL when has idle core
To:     Abel Wu <wuyun.abel@bytedance.com>
Cc:     Peter Zijlstra <peterz@infradead.org>,
        Mel Gorman <mgorman@suse.de>,
        Vincent Guittot <vincent.guittot@linaro.org>,
        Josh Don <joshdon@google.com>, Chen Yu <yu.c.chen@intel.com>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Jul 12, 2022 at 4:45 PM Abel Wu <wuyun.abel@bytedance.com> wrote:
>
> When SIS_UTIL is enabled, SIS domain scan will be skipped if
> the LLC is overloaded. Since the overloaded status is checked
> in the load balancing at LLC level, the interval is llc_size
> miliseconds. The duration might be long enough to affect the
> overall system throughput if idle cores are out of reach in
> SIS domain scan.
>
> Signed-off-by: Abel Wu <wuyun.abel@bytedance.com>
> ---
>
Tested schbench and netperf on latest 5.19 vanilla, it seems that there
is latency performance improvement when the load is low in schbench,
and no performance difference on netperf.

 ./report.py -b 5.19.0+ -c 5.19.0-skip-sis-util+ -t schbench

schbench
========
case            load    baseline(std%) compare%( std%)
normal          mthread-1 1.00 (  0.00) +7.69 (  0.00)
normal          mthread-2 1.00 (  0.00) +13.24 (  0.00)
normal          mthread-4 1.00 (  0.00) -5.88 (  0.00)
normal          mthread-8 1.00 (  0.00) -0.25 (  0.00)


./report.py -b 5.19.0+ -c 5.19.0-skip-sis-util+ -t netperf
netperf
=======
case            load    baseline(std%) compare%( std%)
TCP_RR          thread-28 1.00 (  0.62) +0.15 (  0.55)
TCP_RR          thread-56 1.00 (  0.42) -0.26 (  0.40)
TCP_RR          thread-84 1.00 (  0.29) +0.39 (  0.29)
TCP_RR          thread-112 1.00 (  0.22) +0.44 (  0.23)
TCP_RR          thread-140 1.00 (  0.17) +0.33 (  0.18)
TCP_RR          thread-168 1.00 (  0.17) +0.19 (  0.16)
TCP_RR          thread-196 1.00 ( 13.65) -0.62 ( 14.83)
TCP_RR          thread-224 1.00 (  9.80) -0.65 (  9.67)
UDP_RR          thread-28 1.00 (  0.89) +0.92 (  0.81)
UDP_RR          thread-56 1.00 (  0.78) +0.38 (  0.73)
UDP_RR          thread-84 1.00 ( 14.03) +0.78 ( 16.85)
UDP_RR          thread-112 1.00 ( 12.26) -0.42 ( 11.95)
UDP_RR          thread-140 1.00 (  9.86) -0.89 (  6.93)
UDP_RR          thread-168 1.00 ( 11.62) -0.82 (  8.80)
UDP_RR          thread-196 1.00 ( 19.47) +0.42 ( 16.50)
UDP_RR          thread-224 1.00 ( 18.68) +0.72 ( 18.50)


Tested-by: Chen Yu <yu.c.chen@intel.com>

-- 
Thanks,
Chenyu