From owner-freebsd-stable@FreeBSD.ORG Mon Oct 23 17:37:53 2006 Return-Path: X-Original-To: freebsd-stable@freebsd.org Delivered-To: freebsd-stable@freebsd.org Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id EA8F816A582; Mon, 23 Oct 2006 17:37:53 +0000 (UTC) (envelope-from mi+mx@aldan.algebra.com) Received: from corbulon.video-collage.com (static-151-204-231-237.bos.east.verizon.net [151.204.231.237]) by mx1.FreeBSD.org (Postfix) with ESMTP id 1028443D75; Mon, 23 Oct 2006 17:37:40 +0000 (GMT) (envelope-from mi+mx@aldan.algebra.com) Received: from [172.21.130.86] (mx-broadway [38.98.68.18]) by corbulon.video-collage.com (8.13.6/8.13.6) with ESMTP id k9NHbdkF055590 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NO); Mon, 23 Oct 2006 13:37:39 -0400 (EDT) (envelope-from mi+mx@aldan.algebra.com) From: Mikhail Teterin Organization: Virtual Estates, Inc. To: Gleb Smirnoff Date: Mon, 23 Oct 2006 13:37:33 -0400 User-Agent: KMail/1.9.1 References: <200610211300.09476@aldan> <20061021173358.GC75694@cell.sick.ru> In-Reply-To: <20061021173358.GC75694@cell.sick.ru> MIME-Version: 1.0 Content-Type: text/plain; charset="koi8-r" Content-Transfer-Encoding: 8bit Content-Disposition: inline Message-Id: <200610231337.33873.mi+mx@aldan.algebra.com> X-Virus-Scanned: ClamAV 0.88.4/2082/Mon Oct 23 10:58:17 2006 on corbulon.video-collage.com X-Virus-Status: Clean X-Scanned-By: MIMEDefang 2.43 Cc: freebsd-net@freebsd.org, freebsd-stable@freebsd.org Subject: Re: em network issues X-BeenThere: freebsd-stable@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: Production branch of FreeBSD source code List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Mon, 23 Oct 2006 17:37:54 -0000 ΣΥΒΟΤΑ 21 ΦΟΧΤΕΞΨ 2006 13:33, Gleb Smirnoff ΞΑΠΙΣΑΧ: > We aren't currently speaking about performance, we need to know whether > kernel with DEVICE_POLLING option makes NIC work stable. Yes, that seems to be the case... After I got to the machine's console (there was no network access) and turning off the polling on em0, it started accepting the dumps (two streams in parallel from two different remote servers), and compressing them (zlib -9) at about 35 Mb/s. The em0's interrupts fire between 3K and 4K times per second, accounting for 3-7% of the total CPU load. I'll test with three parallel dumps tonight (one of them from an "overwhelming" 8-Sparc db-server), but it looks quite promising already. Yours, -mi 4 users Load 2.46 2.34 1.63 23 ΦΟΧ 13:34 Mem:KB REAL VIRTUAL VN PAGER SWAP PAGER Tot Share Tot Share Free in out in out Act 113456 17640 233076 44512 89204 count All 2001588 20244 1426659k 51588 pages Interrupts Proc:r p d s w Csw Trp Sys Int Sof Flt cow 7738 total 2 60 12703 172391928576 70 254372 wire irq1: atkb 120004 act irq6: fdc0 4.9%Sys 6.3%Intr 88.8%User 0.0%Nice 0.0%Idl 1548080 inact irq15: ata | | | | | | | | | | 86240 cache irq17: fwo ==++++>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> 2964 free irq20: nve daefr irq21: ohc Namei Name-cache Dir-cache prcfr 2 irq22: ehc Calls hits % hits % react 3679 irq25: em0 312 312 100 pdwak 67 irq29: amr zfod pdpgs 1995 cpu0: time Disks ad4 ad6 amrd0 ozfod intrn 1995 cpu1: time KB/t 128 0.00 75.03 %slo-z 221184 buf tps 1 0 58 1112 tfree 23 dirtybuf MB/s 0.12 0.00 4.23 100000 desiredvnodes % busy 0 0 61 2757 numvnodes Showing vmstat, refresh every 1 seconds. 1965