[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-Id: <19FD6A47-11AA-4C76-B5B6-98BB84971D16@coly.li>
Date: Mon, 26 May 2025 17:44:31 +0800
From: Coly Li <i@...y.li>
To: Zhou Jifeng <zhoujifeng@...inos.com.cn>
Cc: Coly Li <colyli@...nel.org>,
邓旺波 <dengwangbo@...inos.com.cn>,
"kent.overstreet" <kent.overstreet@...ux.dev>,
linux-bcache <linux-bcache@...r.kernel.org>,
linux-kernel <linux-kernel@...r.kernel.org>,
夏华 <xiahua@...inos.com.cn>
Subject: Re: [PATCH v2] bcache: add the deferred_flush IO processing path in
the writeback mode
> 2025年5月26日 17:42,Zhou Jifeng <zhoujifeng@...inos.com.cn> 写道:
>
> On Mon, 26 May 2025 at 15:42, Coly Li <i@...y.li> wrote:
>>
>> Hi Jifeng,
>>
>>> 2025年5月26日 14:41,Zhou Jifeng <zhoujifeng@...inos.com.cn> 写道:
>>>
>>> On Mon, 28 Apr 2025 at 15:36, Zhou Jifeng <zhoujifeng@...inos.com.cn> wrote:
>>> .....
>>>> v1->v2: Version v2 mainly addresses the issue of low efficiency in
>>>> writing back dirty data in version v1. When writing back dirty data,
>>>> it no longer uses the FUA method but instead writes back no more than
>>>> 500 dirty bkeys and then uniformly sends a PREFLUSH instruction once.
>>>> I will supplement more test data later.
>>> .....
>>>
>>> Comparison test data::
>>> https://github.com/jifengzhou/file/blob/main/bcache-deferred-patch-correlation-data.pdf
>>>
>>
>> Can I access the raw data to have a look?
>>
>> And the three different testings, which parameters of bcache are modified from default?
>>
>> Thank.
>>
>>
>> Coly Li
>
> Test raw data address:
> https://github.com/jifengzhou/file/tree/main/deferred%20test%20data
>
> I have not learned the default values of the parameters in the configuration file. Generally, they are
> configured according to the required characteristics, such as random IO, sequential IO, block size, etc.
> The only difference between the first and second test parameters is openflags. The difference between
> the third test and the first two groups is openflags, xfersize, and seekpct. The xfersize parameter is
> used to control the access block size, and the seekpct parameter is used to control the random ratio. 0
> represents complete order and 100 represents 100% random.
Copied. Because the data lines from chats are too closed, I need to read them more close.
Thanks.
Coly Li
Powered by blists - more mailing lists