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;
Subject: Re: [PATCH -next RFC 2/6] block: refactor to split bio thoroughly
To:     Jens Axboe <axboe@kernel.dk>, Christoph Hellwig <hch@infradead.org>
CC:     <andriy.shevchenko@linux.intel.com>, <john.garry@huawei.com>,
        <ming.lei@redhat.com>, <linux-block@vger.kernel.org>,
        <linux-kernel@vger.kernel.org>, <yi.zhang@huawei.com>
References: <20220329094048.2107094-1-yukuai3@huawei.com>
 <20220329094048.2107094-3-yukuai3@huawei.com>
 <YkMKgwsZ3K8dRVbX@infradead.org>
 <e46db33f-84a6-6d23-9be2-7429fec71046@kernel.dk>
 <YkMabVKr+7Mg53E1@infradead.org>
 <338207c2-e3cd-abc1-6b8c-f08645ef144a@kernel.dk>
From:   "yukuai (C)" <yukuai3@huawei.com>
Message-ID: <fe6c3d68-3f44-0bf0-b43b-64f57cf83f9c@huawei.com>
Date:   Wed, 30 Mar 2022 09:54:03 +0800
User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:60.0) Gecko/20100101
 Thunderbird/60.8.0
MIME-Version: 1.0
In-Reply-To: <338207c2-e3cd-abc1-6b8c-f08645ef144a@kernel.dk>
Content-Type: text/plain; charset="utf-8"; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 2022/03/29 22:41, Jens Axboe wrote:
> On 3/29/22 8:40 AM, Christoph Hellwig wrote:
>> On Tue, Mar 29, 2022 at 08:35:29AM -0600, Jens Axboe wrote:
>>>> But more importantly why does your use case even have splits that get
>>>> submitted together?  Is this a case of Linus' stupidly low default
>>>> max_sectors when the hardware supports more, or is the hardware limited
>>>> to a low number of sectors per request?  Or do we hit another reason
>>>> for the split?
>>>
>>> See the posted use case, it's running 512kb ios on a 128kb device.
Hi,

The problem was first found during kernel upgrade(v3.10 to v4.18), and
we maintain a series of io performance test suites, and one of the test
is fio random rw with large bs. In the environment, the 'max_sectors_kb'
is 256kb, and fio bs is 1m.
>>
>> That is an awfully low limit these days.  I'm really not sure we should
>> optimize the block layer for that.
> 
> That's exactly what my replies have been saying. I don't think this is
> a relevant thing to optimize for.

If the use case that large ios get submitted together is not a common
issue（probably not since it's been a long time without complaining),
I agree that we should not optimize the block layer for that.

Thanks,
Kuai
> 
> Fixing fairness for wakeups seems useful, however.
>