From owner-freebsd-fs@FreeBSD.ORG Wed Dec 16 05:05:28 2009 Return-Path: Delivered-To: freebsd-fs@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:4f8:fff6::34]) by hub.freebsd.org (Postfix) with ESMTP id 15E5D1065672 for ; Wed, 16 Dec 2009 05:05:28 +0000 (UTC) (envelope-from jonathan@kc8onw.net) Received: from mail.kc8onw.net (kc8onw.net [206.55.209.81]) by mx1.freebsd.org (Postfix) with ESMTP id CEED78FC0A for ; Wed, 16 Dec 2009 05:05:27 +0000 (UTC) Received: from [10.70.3.199] (c-98-226-147-124.hsd1.in.comcast.net [98.226.147.124]) by mail.kc8onw.net (Postfix) with ESMTPSA id E01FB1E134; Tue, 15 Dec 2009 23:49:30 -0500 (EST) Message-ID: <4B2866A5.3080207@kc8onw.net> Date: Tue, 15 Dec 2009 23:48:37 -0500 From: Jonathan User-Agent: Mozilla/5.0 (Windows; U; Windows NT 6.0; en-US; rv:1.9.1.5) Gecko/20091204 Thunderbird/3.0 MIME-Version: 1.0 To: ivoras@gmail.com, freebsd-fs@freebsd.org Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: Subject: RE: ZFS, compression, system load, pauses (livelocks?) X-BeenThere: freebsd-fs@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: Filesystems List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Wed, 16 Dec 2009 05:05:28 -0000 I seem to have run into the same problem with "2) long pauses, in what looks like vfs.zfs.txg.timeout second intervals" http://lists.freebsd.org/pipermail/freebsd-fs/2009-December/007343.html In my case 50-100% CPU is used by ZFS with *no* disk activity during the pauses then a burst of rapid disk activity and then another pause. I'm also not running compression on the file system that I am writing to so I don't think it's something specific to compression. Has anyone had any luck finding a solution or are people still just patching around it for now? I dropped vfs.zfs.txg.timeout from 30 to 5 seconds and my throughput is far better, but still sawtoothed. The actual data transfer "teeth" are much closer together but still seem to be spaced at vfs.zfs.txg.timeout intervals. When transferring data I see about 50% of a 1gb link which drops to 0 during the pauses. Based on gstat my disks spend maybe 1/4 of their time busy so I doubt my array is the limiting factor in this situation. I'm running 8-stable r200414 right now and I don't remember having this problem with 8-beta releases so maybe something has changed recently that triggered this? Jonathan Stewart Sorry for the broken threading. I've added freebsd-fs to my subscription list so I will be able to follow the rest of the discussion on the list.