From owner-freebsd-fs@FreeBSD.ORG Sat Sep 17 18:14:55 2011 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 02546106566C for ; Sat, 17 Sep 2011 18:14:55 +0000 (UTC) (envelope-from lytboris@gmail.com) Received: from mail-fx0-f54.google.com (mail-fx0-f54.google.com [209.85.161.54]) by mx1.freebsd.org (Postfix) with ESMTP id 8F5A28FC0A for ; Sat, 17 Sep 2011 18:14:53 +0000 (UTC) Received: by fxg9 with SMTP id 9so3624875fxg.13 for ; Sat, 17 Sep 2011 11:14:52 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :cc:content-type; bh=D7pX2ZciewCoDqC/df7VmKv4UcMsVm9AX9JR2B9Coxo=; b=nfonVJa/D1rSDBnq5pHeJCo4vwYd1auDk3vWi07QZrf/BzIE3+CeMUdtqs2JKZhf79 0aSt+qmSVpdNM904EMpcDyX5KtFlRpRomq0sUYgyskYFjrh4X8U+aa4AWIzWmk9U5rA3 Lkqy0gg+q+Wlk8l6Lv3aRhCMPmsaRWKEuCyCk= MIME-Version: 1.0 Received: by 10.223.17.3 with SMTP id q3mr1593897faa.71.1316283292674; Sat, 17 Sep 2011 11:14:52 -0700 (PDT) Received: by 10.223.106.15 with HTTP; Sat, 17 Sep 2011 11:14:52 -0700 (PDT) In-Reply-To: <6B437FA4-B422-4BE7-BDF5-F90717F3865B@bitgravity.com> References: <6B437FA4-B422-4BE7-BDF5-F90717F3865B@bitgravity.com> Date: Sat, 17 Sep 2011 22:14:52 +0400 Message-ID: From: Lytochkin Boris To: "David P. Discher" Content-Type: text/plain; charset=ISO-8859-1 Cc: freebsd-fs@freebsd.org Subject: Re: [ZFS] starving reads while idle disks 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: Sat, 17 Sep 2011 18:14:55 -0000 Hi. > Do you see the same read-stravation when writing the tar to a file ? (possibly outside the zpool). Yep. > I have anecdotal suspicion that /dev/null has some performance hit of blocking or locking. No-no. Every program that tries to read from ZFS faces this issue actually. I found something more interesting. Lets presume I have 10 big dirs in . Issuing tar|dd command on separate dirs (so spawning 10 "threads") simultaneously will result in 10x faster reads cumulatively (I saw 15Mb/s in zpool iostat), and disk load may be as high as 70%. So I think there is some read throttling thing that limits read speed per read(). But still no clue where to find this bottleneck. -- Wbr, Boris.