Received-SPF: pass (google.com: domain of linux-ext4-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <2217188c-7009-f142-3e7e-b3e61d2c1324@i2se.com>
Date:   Fri, 29 Jul 2022 07:30:25 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.9.1
Subject: Re: [Regression] ext4: changes to mb_optimize_scan cause issues on
 Raspberry Pi
Content-Language: en-US
To:     Jan Kara <jack@suse.cz>
Cc:     linux-ext4@vger.kernel.org, Ojaswin Mujoo <ojaswin@linux.ibm.com>,
        Harshad Shirwadkar <harshadshirwadkar@gmail.com>,
        Theodore Ts'o <tytso@mit.edu>,
        Ritesh Harjani <riteshh@linux.ibm.com>,
        linux-fsdevel@vger.kernel.org,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Geetika.Moolchandani1@ibm.com, regressions@lists.linux.dev
References: <0d81a7c2-46b7-6010-62a4-3e6cfc1628d6@i2se.com>
 <20220728100055.efbvaudwp3ofolpi@quack3>
From:   Stefan Wahren <stefan.wahren@i2se.com>
In-Reply-To: <20220728100055.efbvaudwp3ofolpi@quack3>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

Hi Jan,

Am 28.07.22 um 12:00 schrieb Jan Kara:
> Hello!
> Can you run "iostat -x 1" while the download is running so that we can see
> roughly how the IO pattern looks?
>
> Also can get filesystem metadata image of your card like:
>    e2image -r <fs-device> - | gzip >/tmp/ext4-image.gz
>
> and put it somewhere for download? The image will contain only fs metadata,
> not data so it should be relatively small and we won't have access to your
> secrets ;). With the image we'd be able to see how the free space looks
> like and whether it perhaps does not trigger some pathological behavior.
>
> My current suspicion is that because the new allocator strategy spreads
> allocations over more block groups, we end up with more open erase blocks
> on the SD card which forces the firmware to do more garbage collection and
> RMW of erase blocks and write performance tanks...
>
> Thanks.
thanks for your feedback. Unfortunately i'm busy the next days, so it 
will take some time to provide this.
> 								Honza
>