From mboxrd@z Thu Jan 1 00:00:00 1970 From: Rick Jones Subject: Re: Latency difference between fifo and pfifo_fast Date: Tue, 06 Dec 2011 10:20:57 -0800 Message-ID: <4EDE5D09.1060404@hp.com> References: <03369d06-ee52-4eef-a8b6-5a21e6d56840@jasiiieee> <1323161496.2448.7.camel@edumazet-HP-Compaq-6005-Pro-SFF-PC> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Eric Dumazet , netdev@vger.kernel.org To: "John A. Sullivan III" Return-path: Received: from g4t0014.houston.hp.com ([15.201.24.17]:2712 "EHLO g4t0014.houston.hp.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752735Ab1LFSVB (ORCPT ); Tue, 6 Dec 2011 13:21:01 -0500 In-Reply-To: <1323161496.2448.7.camel@edumazet-HP-Compaq-6005-Pro-SFF-PC> Sender: netdev-owner@vger.kernel.org List-ID: On 12/06/2011 12:51 AM, Eric Dumazet wrote: > Le mardi 06 d=C3=A9cembre 2011 =C3=A0 03:39 -0500, John A. Sullivan I= II a =C3=A9crit : > >>> ifconfig eth2 txqueuelen 0 >>> tc qdisc add dev eth2 root pfifo >>> tc qdisc del dev eth2 root >>> >>> >>> >>> >> Really? I didn't know one could do that. Thanks. However, with no >> queue length, do I have a significant risk of dropping packets? To >> answer your other response's question, these are Intel quad port e10= 00 >> cards. We are frequently pushing them to near line speed so >> 1,000,000,000 / 1534 / 8 =3D 81,486 pps - John > > You can remove qdisc layer, since NIC itself has a TX ring queue > > (check exact value with ethtool -g ethX) > > # ethtool -g eth2 > Ring parameters for eth2: > Pre-set maximums: > RX: 4078 > RX Mini: 0 > RX Jumbo: 0 > TX: 4078 > Current hardware settings: > RX: 254 > RX Mini: 0 > RX Jumbo: 0 > TX: 4078 ---- HERE ---- And while you are down at the NIC, if every microsecond is precious (no= =20 matter how close to epsilon compared to the latencies of spinning rust=20 :) you might consider disabling interrupt coalescing via ethtool -C. rick jones