[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CANn89i+AS+YE1w9E9LxKsGtqgXW7uKjZv==MYFx9UwPSe-EM1g@mail.gmail.com>
Date: Mon, 23 Dec 2019 12:22:03 -0800
From: Eric Dumazet <edumazet@...gle.com>
To: "David S . Miller" <davem@...emloft.net>
Cc: netdev <netdev@...r.kernel.org>,
Eric Dumazet <eric.dumazet@...il.com>,
Soheil Hassas Yeganeh <soheil@...gle.com>,
Neal Cardwell <ncardwell@...gle.com>,
Yuchung Cheng <ycheng@...gle.com>,
Martin KaFai Lau <kafai@...com>
Subject: Re: [PATCH net-next 5/5] tcp_cubic: make Hystart aware of pacing
On Mon, Dec 23, 2019 at 12:20 PM Eric Dumazet <edumazet@...gle.com> wrote:
>
> For years we disabled Hystart ACK train detection at Google
> because it was fooled by TCP pacing.
>
> ACK train detection uses a simple heuristic, detecting if
> we receive ACK past half the RTT, to exit slow start before
> hitting the bottleneck and experience massive drops.
>
>
...
> + threshold = ca->delay_min;
> + /* Hystart ack train triggers if we get ack past
> + * ca->delay_min/2.
> + * Pacing might have delayed packets up to RTT/2
> + * during slow start.
> + */
> + if ((sk->sk_pacing_status == SK_PACING_NONE) || (netdev_max_backlog & 1))
> + threshold >>= 1;
This is silly, last minute debug feature that I forgot to remove
Will send a V2 without it.
Powered by blists - more mailing lists