Received-SPF: softfail (google.com: domain of transitioning linux-kernel-owner@vger.kernel.org does not designate 23.128.96.19 as permitted sender) client-ip=23.128.96.19;
Message-ID: <c14c0103-9cbd-7d0f-486b-344dd33725ab@deltatee.com>
Date:   Thu, 21 Apr 2022 10:02:33 -0600
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.7.0
Content-Language: en-CA
To:     Xiao Ni <xni@redhat.com>
Cc:     open list <linux-kernel@vger.kernel.org>,
        linux-raid <linux-raid@vger.kernel.org>,
        Song Liu <song@kernel.org>,
        Christoph Hellwig <hch@infradead.org>,
        Guoqing Jiang <guoqing.jiang@linux.dev>,
        Stephen Bates <sbates@raithlin.com>,
        Martin Oliveira <Martin.Oliveira@eideticom.com>,
        David Sloan <David.Sloan@eideticom.com>
References: <20220420195425.34911-1-logang@deltatee.com>
 <CALTww28fwNpm0O_jc7-2Xr0JSX9i6F1kgoUQ8m_k6ZgPa1XxXw@mail.gmail.com>
From:   Logan Gunthorpe <logang@deltatee.com>
In-Reply-To: <CALTww28fwNpm0O_jc7-2Xr0JSX9i6F1kgoUQ8m_k6ZgPa1XxXw@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Subject: Re: [PATCH v2 00/12] Improve Raid5 Lock Contention
Precedence: bulk


On 2022-04-21 02:45, Xiao Ni wrote:
> Could you share the commands to get the test result (lock contention
> and performance)?

Sure. The performance we were focused on was large block writes. So we
setup raid5 instances with varying number of disks and ran the following
fio script directly on the drive.

[simple]
filename=/dev/md0
ioengine=libaio
rw=write
direct=1
size=8G
blocksize=2m
iodepth=16
runtime=30s
time_based=1
offset_increment=8G
numjobs=12
￼
(We also played around with tuning this but didn't find substantial
changes once the bottleneck was hit)

We tuned md with parameters like:

echo 4 > /sys/block/md0/md/group_thread_cnt
echo 8192 > /sys/block/md0/md/stripe_cache_size

For lock contention stats, we just used lockstat[1]; roughly like:

echo 1 > /proc/sys/kernel/lock_stat
fio test.fio
echo 0 > /proc/sys/kernel/lock_stat
cat /proc/lock_stat

And compared the before and after.

Logan

[1] https://www.kernel.org/doc/html/latest/locking/lockstat.html