From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from ns.mahan.org (unknown [67.116.10.138]) by dpdk.org (Postfix) with ESMTP id E12DE6938 for ; Wed, 10 Jul 2013 22:18:34 +0200 (CEST) Received: from [192.168.71.25] (localhost [127.0.0.1]) (authenticated bits=0) by ns.mahan.org (8.14.5/8.14.5) with ESMTP id r6AKIXsJ038193 for ; Wed, 10 Jul 2013 13:18:34 -0700 (PDT) (envelope-from mahan@mahan.org) From: Patrick Mahan Content-Type: text/plain; charset=us-ascii X-Mailer: iPad Mail (10B329) Message-Id: <324D7C36-DE37-44EB-A3D8-DB3D2DA2E8AB@mahan.org> Date: Wed, 10 Jul 2013 13:18:32 -0700 To: dev@dpdk.org Content-Transfer-Encoding: quoted-printable Mime-Version: 1.0 (1.0) Subject: [dpdk-dev] TX write threshold vs Report Status threshold X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: patches and discussions about DPDK List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Wed, 10 Jul 2013 20:18:35 -0000 I'm not sure this is a bug or feature. First, I've been away for a few weeks being on vacation and other stuff. My= client liked my findings on DPDK well enough that I have been loaned a dual= -cpu system (2 Xeon E5-2690 @ 2.9 GHz) and two 82599 cards. This is so I ca= n eliminate the PCIe bus contention and to test the Numa capabilities.=20 So I am now seeing (almost) line rate (~14.32 Mpps) between the two cards. B= ut it is not quite line rate (14.88 Mpps), so I have started playing around w= ith the TX/RX threshold parameters. Reading the datasheet for the 82599, section 7.2.3.5.1 seems to indicate tha= t the write threshold (TXDCTL[n].WTHRESH) should (could?) set this value to o= btain more efficient use of the PCIe bus and memory. However, testpmd fails because the TX rs threshold (set to 0 by testpmd) is d= efaulting to 32 inside the ixgbe pmd code, which then fails because I have s= et the TX write threshold to 16. And I cannot force the TX rs threshold < 0 via the command line. Personally I think this is a bug, but maybe it's a feature :-) but is there a= reason to force use of the rs threshold over the write threshold? Thanks, Patrick