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;
MIME-Version: 1.0
References: <20221229080207.1029-1-cuiyunhui@bytedance.com> <Y7HaeTkNtfb3oIP4@pop-os.localdomain>
In-Reply-To: <Y7HaeTkNtfb3oIP4@pop-os.localdomain>
From:   =?UTF-8?B?6L+Q6L6J5bSU?= <cuiyunhui@bytedance.com>
Date:   Wed, 4 Jan 2023 11:34:53 +0800
Message-ID: <CAEEQ3wmc_hPwUc9zeFJ7c0xvmyip7CGHvCHQN2U4c8wyfM1KLA@mail.gmail.com>
Subject: Re: [External] Re: [PATCH] tcp/udp: add tracepoint for send recv length
To:     Cong Wang <xiyou.wangcong@gmail.com>
Cc:     edumazet@google.com, rostedt@goodmis.org, mhiramat@kernel.org,
        davem@davemloft.net, yoshfuji@linux-ipv6.org, dsahern@kernel.org,
        kuba@kernel.org, pabeni@redhat.com, duanxiongchun@bytedance.com,
        netdev@vger.kernel.org, linux-kernel@vger.kernel.org,
        linux-trace-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

Cong Wang <xiyou.wangcong@gmail.com> =E4=BA=8E2023=E5=B9=B41=E6=9C=882=E6=
=97=A5=E5=91=A8=E4=B8=80 03:10=E5=86=99=E9=81=93=EF=BC=9A
>
> On Thu, Dec 29, 2022 at 04:02:07PM +0800, Yunhui Cui wrote:
> > From: Xiongchun Duan <duanxiongchun@bytedance.com>
> >
> > Add a tracepoint for capturing TCP segments with
> > a send or receive length. This makes it easy to obtain
> > the packet sending and receiving information of each process
> > in the user mode, such as the netatop tool.
>
> You can obtain the same information with kretprobe:
> https://www.gcardone.net/2020-07-31-per-process-bandwidth-monitoring-on-L=
inux-with-bpftrace/

As we know, kprobe gets the result by trapping in an exception, which
loses performance compared to tracepoint.

We did a test for performance comparison. The results are as follows.

Time per request
sock_sendmsg(k,kr):  12.382ms=EF=BC=8C tcp_send_length(tracepoint): 11.887m=
s=EF=BC=8C
without hook=EF=BC=9A11.222ms

It can be seen that the performance loss of tracepoint is only half of
that of kprobe.