From owner-freebsd-fs@FreeBSD.ORG Tue Oct 19 11:54:04 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 AA011106564A; Tue, 19 Oct 2010 11:54:04 +0000 (UTC) (envelope-from petefrench@ticketswitch.com) Received: from constantine.ticketswitch.com (constantine.ticketswitch.com [IPv6:2002:57e0:1d4e:1::3]) by mx1.freebsd.org (Postfix) with ESMTP id 6DF278FC1F; Tue, 19 Oct 2010 11:54:04 +0000 (UTC) Received: from dilbert.rattatosk ([10.64.50.6] helo=dilbert.ticketswitch.com) by constantine.ticketswitch.com with esmtps (TLSv1:AES256-SHA:256) (Exim 4.69 (FreeBSD)) (envelope-from ) id 1P8AlX-000Hid-LF; Tue, 19 Oct 2010 12:54:03 +0100 Received: from petefrench by dilbert.ticketswitch.com with local (Exim 4.72 (FreeBSD)) (envelope-from ) id 1P8AlX-0001P4-KI; Tue, 19 Oct 2010 12:54:03 +0100 Date: Tue, 19 Oct 2010 12:54:03 +0100 Message-Id: To: pjd@freebsd.org, zbeeble@gmail.com In-Reply-To: From: Pete French Cc: freebsd-fs@freebsd.org Subject: Re: Converting a non-HAST ZFS pool to a HAST pool 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: Tue, 19 Oct 2010 11:54:04 -0000 > I'm wondering if I'm missing something here --- because I'm wondering > if running HAST under ZFS isn't a step backwards. > > My quick read of HAST seems to indicate that it's going to manage two > disks and present them as one disk to ZFS. The design problem with > this (especially since we're talking a _lot_ of network (and memory) > transfers involved) is data corruption --- the idea that ZFS protects > data better when it can determine one disk has it right while another > disk has it wrong (as it can when it manages the two disks). > > Wouldn't it be better to just have network (iscsi-like) spools > attached to ZFS? Individual spools could still fail. What am I > missing? Is there a better description of HAST than the FreeBSD wiki > page? I guess in theory the answer there is 'yes' - but have you actually tried it in practice ? I did this for a while as an experiment using ggated - the problem is that when the remote fails (for example) it doesnt signal up to ZFS properly and instead of ZFS seeing a failed driev it just locks up. iSCSI had similar issues. Note that this test was a while ago, and the situation may have improved - but that was the way I did it at the time I set this up. Interstingly, I am considering a hybrid - using ZFS as a mirror over a pair of HAST devices. My servers have hardwar RAID in them and a pair of drives each, so I could split those, remove the hardware RAID, and run 4 drives as 2 hast devices with ZFS on top. That soulds like the best soultuon to me - but you do then have the issue of doubling your network bandwidth required. -pete.