Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [RFC][PATCH 03/16] sched: Wrap rq::lock access
From:   Subhra Mazumdar <subhra.mazumdar@oracle.com>
To:     Julien Desfossez <jdesfossez@digitalocean.com>
Cc:     Peter Zijlstra <peterz@infradead.org>, mingo@kernel.org,
        tglx@linutronix.de, pjt@google.com, tim.c.chen@linux.intel.com,
        torvalds@linux-foundation.org, linux-kernel@vger.kernel.org,
        fweisbec@gmail.com, keescook@chromium.org, kerrnel@google.com,
        Vineeth Pillai <vpillai@digitalocean.com>,
        Nishanth Aravamudan <naravamudan@digitalocean.com>
References: <fcad329b-e421-624a-0d1a-a98466eaf2ed@oracle.com>
 <1553866527-18879-1-git-send-email-jdesfossez@digitalocean.com>
 <6e8e6fa0-8976-5e97-d90c-af0b4a6fc8b2@oracle.com>
Message-ID: <cd11cde9-4aed-b02f-cae5-170bb6fa686b@oracle.com>
Date:   Mon, 1 Apr 2019 14:35:51 -0700
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.5.1
MIME-Version: 1.0
In-Reply-To: <6e8e6fa0-8976-5e97-d90c-af0b4a6fc8b2@oracle.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
Content-Language: en-US
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


On 3/29/19 3:23 PM, Subhra Mazumdar wrote:
>
> On 3/29/19 6:35 AM, Julien Desfossez wrote:
>> On Fri, Mar 22, 2019 at 8:09 PM Subhra Mazumdar 
>> <subhra.mazumdar@oracle.com>
>> wrote:
>>> Is the core wide lock primarily responsible for the regression? I ran
>>> upto patch
>>> 12 which also has the core wide lock for tagged cgroups and also calls
>>> newidle_balance() from pick_next_task(). I don't see any 
>>> regression.  Of
>>> course
>>> the core sched version of pick_next_task() may be doing more but
>>> comparing with
>>> the __pick_next_task() it doesn't look too horrible.
>> On further testing and investigation, we also agree that spinlock 
>> contention
>> is not the major cause for the regression, but we feel that it should 
>> be one
>> of the major contributing factors to this performance loss.
>>
>>
> I finally did some code bisection and found the following lines are
> basically responsible for the regression. Commenting them out I don't see
> the regressions. Can you confirm? I am yet to figure if this is needed 
> for
> the correctness of core scheduling and if so can we do this better?
>
> -------->8-------------
>
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index fe3918c..3b3388a 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -3741,8 +3741,8 @@ pick_next_task(struct rq *rq, struct task_struct 
> *prev, struct rq_flags *rf)
>                                  * If there weren't no cookies; we 
> don't need
>                                  * to bother with the other siblings.
> */
> -                               if (i == cpu && !rq->core->core_cookie)
> -                                       goto next_class;
> +                               //if (i == cpu && !rq->core->core_cookie)
> +                                       //goto next_class;
>
> continue;
>                         }
AFAICT this condition is not needed for correctness as cookie matching will
sill be enforced. Peter any thoughts? I get the following numbers with 1 DB
and 2 DB instance.

1 DB instance
users  baseline   %idle  core_sched %idle
16     1          84     -5.5% 84
24     1          76     -5% 76
32     1          69     -0.45% 69

2 DB instance
users  baseline   %idle  core_sched %idle
16     1          66     -23.8% 69
24     1          54     -3.1% 57
32     1          42     -21.1%      48