From owner-freebsd-fs@FreeBSD.ORG Fri Jan 29 16:26:10 2010 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 637291065670 for ; Fri, 29 Jan 2010 16:26:10 +0000 (UTC) (envelope-from korvus@comcast.net) Received: from mx04.pub.collaborativefusion.com (mx04.pub.collaborativefusion.com [206.210.72.84]) by mx1.freebsd.org (Postfix) with ESMTP id 2DFC98FC12 for ; Fri, 29 Jan 2010 16:26:09 +0000 (UTC) Received: from reflector.ws.pitbpa0.priv.collaborativefusion.com ([206.210.89.202]) by mx04.pub.collaborativefusion.com (StrongMail Enterprise 4.1.1.4(4.1.1.4-47689)); Fri, 29 Jan 2010 11:35:54 -0500 X-VirtualServerGroup: Default X-MailingID: 00000::00000::00000::00000::::671 X-SMHeaderMap: mid="X-MailingID" X-Destination-ID: freebsd-fs@freebsd.org X-SMFBL: ZnJlZWJzZC1mc0BmcmVlYnNkLm9yZw== Message-ID: <4B63088F.9030407@comcast.net> Date: Fri, 29 Jan 2010 11:10:55 -0500 From: Steve Polyack User-Agent: Mozilla/5.0 (X11; U; FreeBSD amd64; en-US; rv:1.9.1.7) Gecko/20100129 Thunderbird/3.0.1 MIME-Version: 1.0 To: Bob Friesenhahn References: <4B619485.8080106@eenet.ee> In-Reply-To: Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: freebsd-fs@freebsd.org Subject: Re: Reading from ZFS mirror 2x slower than expected? 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: Fri, 29 Jan 2010 16:26:10 -0000 On 01/28/10 14:54, Bob Friesenhahn wrote: > On Thu, 28 Jan 2010, Urmas Lett wrote: > >> While reading from simple zfs mirror 'dd if=zfs.test of=/dev/null >> bs=512k' systat -vm shows individual disks only 50% busy: > > How big is the file 'zfs.test'? How was it created? If it was > created from /dev/zero then its content is bogus. You need to test > with non-zero data. > > It may be that zfs's prefetch is not working adequately. With perfect > prefetch, the read rate should be doubled. > I can also vouch that the mirror performance is less than stellar. Writes are also about 50% slower than single-disk performance, although this may be expected due to checksums / other ZFS overhead. I can try to provide more details when I get time. Steve