Message-ID: <ac6a2da7-aa88-b119-6a44-01d2f2ec9b6d@kernel.dk>
Date:   Fri, 10 Mar 2023 13:38:20 -0700
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux aarch64; rv:102.0) Gecko/20100101
 Thunderbird/102.8.0
Subject: Re: [PATCH] io_uring: One wqe per wq
Content-Language: en-US
To:     Breno Leitao <leitao@debian.org>, asml.silence@gmail.com,
        io-uring@vger.kernel.org
Cc:     leit@fb.com, linux-kernel@vger.kernel.org
References: <20230310201107.4020580-1-leitao@debian.org>
From:   Jens Axboe <axboe@kernel.dk>
In-Reply-To: <20230310201107.4020580-1-leitao@debian.org>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 3/10/23 1:11 PM, Breno Leitao wrote:
> Right now io_wq allocates one io_wqe per NUMA node.  As io_wq is now
> bound to a task, the task basically uses only the NUMA local io_wqe, and
> almost never changes NUMA nodes, thus, the other wqes are mostly
> unused.

What if the task gets migrated to a different node? Unless the task
is pinned to a node/cpumask that is local to that node, it will move
around freely.

I'm not a huge fan of the per-node setup, but I think the reasonings
given in this patch are a bit too vague and we need to go a bit
deeper on what a better setup would look like.

-- 
Jens Axboe