Date:   Wed, 8 Feb 2023 14:37:30 -0500
From:   "Theodore Ts'o" <tytso@mit.edu>
To:     Jan Kara <jack@suse.cz>
Cc:     "Bhatnagar, Rishabh" <risbhat@amazon.com>,
        akpm@linux-foundation.org, linux-mm@kvack.org,
        linux-ext4@vger.kernel.org, linux-kernel@vger.kernel.org,
        abuehaze@amazon.com
Subject: Re: EXT4 IOPS degradation between 4.14 and 5.10
Message-ID: <Y+P5+jrsZOjqG+VT@mit.edu>
References: <053b60a6-133e-5d59-0732-464d5160772a@amazon.com>
 <20230126093231.ujn6yaxhexwzizp5@quack3>
 <b948ed49-1bdf-b17e-d03a-7ec8dfdb1afc@amazon.com>
 <20230127121721.lerrb36nhj7gdiwm@quack3>
 <6a0fcca5-b869-ffb7-426b-b49a6782c1c0@amazon.com>
 <20230208140247.rt62xdtriopfdb4o@quack3>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230208140247.rt62xdtriopfdb4o@quack3>
Precedence: bulk

On Wed, Feb 08, 2023 at 03:02:47PM +0100, Jan Kara wrote:
> > I believe its the MySQL database though not so sure.
> 
> Well, in that case I think your MySQL DB is somewhat misconfigured. At
> least as far as we have been consulting MySQL / MariaDB developers
> regarding benchmarking, they suggested we should configure the database to
> use direct IO and increase DB internal buffers instead of relying on
> buffered IO and pagecache behavior. And if your fio job is representative
> of the IO load the DB really creates, I'd agree that that would be a saner
> and likely more performant configuration ;)

Could it possibly be Postgres?  I happen to know that Amazon RDS and
Google Cloud SQL support both MySQL and Postgres, and there are some
optimizations that some of us in the Cloud space have been pursuing
which are much easier because MySQL uses Direct I/O, but unfortunately
Postgres uses buffered I/O and doens't support DIO.  :-(

							- Ted