From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Banerjee, Debabrata" Subject: Re: [PATCH] tcp: Fix RFC reference in comment Date: Tue, 13 Jan 2015 18:17:50 -0500 Message-ID: References: <1421183431-29915-1-git-send-email-dbanerje@akamai.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 8BIT Cc: Yuchung Cheng , David Miller , netdev , "linux-kernel@vger.kernel.org" To: John Heffner Return-path: In-Reply-To: Content-Language: en-US Sender: linux-kernel-owner@vger.kernel.org List-Id: netdev.vger.kernel.org On 1/13/15, 5:01 PM, "John Heffner" wrote: >On Tue, Jan 13, 2015 at 4:42 PM, Banerjee, Debabrata > wrote: >> On 1/13/15, 4:36 PM, "Yuchung Cheng" wrote: >> >>>RFC2861 resets the cwnd like in RFC2581, but the rest of the code >>>implements RFC2861. So I think the current comment is fine. >> >> >> No RFC2861 is an experimental RFC that's implemented in >> tcp_cwnd_application_limited(). RFC2861 Recommends reducing the cwnd by >> averaging the current cwnd and the used cwnd as the new cwnd. >> >> >> RFC2581 4.1 Says to set cwnd to initial cwnd if more than one rto has >> passed since the last send. This is what is implemented in the function >> above. > >Look at the code a little closer -- it's decaying cwnd based on number >of timeouts as described in 2861, not resetting to IW as recommended >in 2581. > > -John You're right it's not RFC2581 I was partially misled by the comment (reset/restart window), but it doesn't appear to be doing what rfc2861 3.2 says either: For i=1 To (tcpnow - T_last)/RTO win = min(cwnd, receiver's declared max window) cwnd = max(win/2, MSS) Versus: u32 restart_cwnd = tcp_init_cwnd(tp, dst); restart_cwnd = min(restart_cwnd, cwnd); while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd) cwnd >>= 1; tp->snd_cwnd = max(cwnd, restart_cwnd); It's not using receiver window, it's using cwnd/init_cwnd, it should at least be using tp->snd_wnd, no?. I stumbled onto this because it looks like tcp_cwnd_application_limited() doesn't execute when it should, because tp->snd_cwnd_stamp is being touched much more often than in rfc2861 3.2. Something seems not right here... -Debabrata