Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <212de77f-6ad1-e012-9b49-8b5cebaded63@gmail.com>
Date:   Mon, 16 May 2022 21:48:58 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.8.1
Subject: Re: [PATCH net-next v3 00/10] UDP/IPv6 refactoring
Content-Language: en-US
To:     Paolo Abeni <pabeni@redhat.com>, netdev@vger.kernel.org,
        "David S . Miller" <davem@davemloft.net>,
        Jakub Kicinski <kuba@kernel.org>
Cc:     David Ahern <dsahern@kernel.org>,
        Eric Dumazet <edumazet@google.com>,
        linux-kernel@vger.kernel.org
References: <cover.1652368648.git.asml.silence@gmail.com>
 <b9025eb4d8a1efefbcd04013cbe8e55e98ef66e1.camel@redhat.com>
From:   Pavel Begunkov <asml.silence@gmail.com>
In-Reply-To: <b9025eb4d8a1efefbcd04013cbe8e55e98ef66e1.camel@redhat.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 5/16/22 14:48, Paolo Abeni wrote:
> Hello,
> 
> On Fri, 2022-05-13 at 16:26 +0100, Pavel Begunkov wrote:
>> Refactor UDP/IPv6 and especially udpv6_sendmsg() paths. The end result looks
>> cleaner than it was before and the series also removes a bunch of instructions
>> and other overhead from the hot path positively affecting performance.
>>
>> Testing over dummy netdev with 16 byte packets yields 2240481 tx/s,
>> comparing to 2203417 tx/s previously, which is around +1.6%
> 
> I personally feel that some patches in this series have a relevant
> chance of introducing functional regressions and e.g. syzbot will not
> help to catch them. That risk is IMHO relevant considered that the
> performance gain here looks quite limited.

I can't say I agree with that. First, I do think the code is much
cleaner having just one block checking corking instead of a couple
of random ifs in different places. Same for sin6. Not to mention
negative line count.

Also, assuming this 1.6% translates to ~0.5-1% with fast NICs, that's
still huge, especially when we get >5GB/s in single core zc tests b/w
servers.

If maintainers are not merging it, I think I'll delay the series until
I get another batch of planned optimisations implemented on top.


> There are a few individual changes that IMHO looks like nice cleanup
> e.g. patch 5, 6, 8, 9 and possibly even patch 1.
> 
> I suggest to reduce the patchset scope to them.

-- 
Pavel Begunkov