From owner-freebsd-questions Sun Apr 21 00:25:38 1996 Return-Path: owner-questions Received: (from root@localhost) by freefall.freebsd.org (8.7.3/8.7.3) id AAA01799 for questions-outgoing; Sun, 21 Apr 1996 00:25:38 -0700 (PDT) Received: from pelican.altadena.net (pelican.altadena.com [206.16.90.21]) by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id AAA01792 for ; Sun, 21 Apr 1996 00:25:35 -0700 (PDT) Received: by pelican.altadena.net (Smail3.1.29.1 #10) id m0uAtWW-0000ReC; Sun, 21 Apr 96 00:25 PDT Message-Id: Date: Sun, 21 Apr 96 00:25 PDT From: pete@pelican.altadena.net (Pete Carah) To: questions@freebsd.org Subject: Re: SCSI RAID controller support? In-Reply-To: <199604180921.DAA17482@shell.aros.net> Sender: owner-questions@freebsd.org X-Loop: FreeBSD.org Precedence: bulk In article <199604180921.DAA17482@shell.aros.net> you write: >Lo and behold, Dave Andersen once said: >> Quite obviously, RAID 0 and RAID 3 have the potential to be >> considerably faster than an ordinary disk. I don't know how they compare >> to software striping as in the ccd, I don't think anyone's done any >> comparisons.:) I've used some of these in graphic design applications, >> and they *really* fly. A RAID level 0 array like the FWB Jackhammer is a >> very pretty piece of equipment, though a tad expensive for most people. :) > As a bit of a followup to this before Terry gets the chance to jump on >me: The RAID level 3 disks perform very well in situations where they >need high stustainable data transfer rates. The seek times, because of >the parity checking, aren't as spectacular as level 0 or as something >like software striping (ccd). In a heavy seeking environment with >multiple users, RAID level 0 will still show performance bonuses, level 3 >less so. On our Sparc 1000 (single or double processor doesn't matter for this test) and a SSA model 100 (this is 30 1gb drives set up as a single 23gb raid-5 filesystem; I don't remember if they are wide or narrow but they are all seagate 31200 models) using the default vxvm raid 5 configuration (whatever that means), we get 1.2mbyte write rates (on 500mb files which is our main application for this thing) with the nvram turned off and 2.2 or so with it turned on. Read comes to about 4.7mbytes either way on the same files... FTP across a 100-base-t runs at the same speed as dd if=/dev/zero of=foo bs=32k count= which I used to make the test files in the first place. And no, vxvm apparently doesn't notice that I'm using /dev/zero and fake out the parity :-) It appears that the whole thing is write-limited by the raid. We also need nfs on the 100baseT but haven't tested that for speed... (and I need to run ttcp to see how the Sun hme card compares with the PCI dec ones; hope it'd be faster :-) Amazing having a 100-base-t and FW scsi on the same rather small S-bus card... -- Pete