From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: questions on NAPI processing latency and dropped network packets Date: Mon, 21 Jan 2008 22:31:09 +0100 Message-ID: <47950F1D.4010508@cosmosbay.com> References: <478654C3.60806@nortel.com> <4794F848.9020402@nortel.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: netdev@vger.kernel.org, linux-kernel@vger.kernel.org To: Chris Friesen Return-path: Received: from gw1.cosmosbay.com ([86.65.150.130]:44135 "EHLO gw1.cosmosbay.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751668AbYAUVbW (ORCPT ); Mon, 21 Jan 2008 16:31:22 -0500 In-Reply-To: <4794F848.9020402@nortel.com> Sender: netdev-owner@vger.kernel.org List-ID: Chris Friesen a =E9crit : > I've done some further digging, and it appears that one of the proble= ms=20 > we may be facing is very high instantaneous traffic rates. >=20 > Instrumentation showed up to 222K packets/sec for short periods (at=20 > least 1.1 ms, possibly longer), although the long-term average is dow= n=20 > around 14-16K packets/sec. Instrumentation done where exactly ? tcpdump on the host will lie, since timestamps are probably going to be= host=20 timestamps, giving the time of RX dequeing, and not time of packets arr= ival=20 given by NIC. >=20 > If I bump the rx ring size up to 4096, we can handle all the packets = and=20 > we still have 44% idle on cpu0 and 27% idle on cpu1. >=20 > Is there anything else we can do to minimize the latency of network=20 > packet processing and avoid having to crank the rx ring size up so hi= gh? >=20 You have some tasks that disable softirqs too long. Sometimes, bumping = RX ring=20 size is OK (but you will still have delays), sometimes it is not an opt= ion,=20 since 4096 is the limit on current hardware.