From owner-freebsd-stable Wed May 29 3:26:28 2002 Delivered-To: freebsd-stable@freebsd.org Received: from mail.allcaps.org (mail.allcaps.org [208.252.245.6]) by hub.freebsd.org (Postfix) with ESMTP id C7B7137B407 for ; Wed, 29 May 2002 03:26:24 -0700 (PDT) Received: by mail.allcaps.org (Postfix, from userid 501) id 6A574152DF; Wed, 29 May 2002 03:26:24 -0700 (PDT) Received: from localhost (localhost [127.0.0.1]) by mail.allcaps.org (Postfix) with ESMTP id 644D8152D6; Wed, 29 May 2002 03:26:24 -0700 (PDT) Date: Wed, 29 May 2002 03:26:24 -0700 (PDT) From: "Andrew P. Lentvorski" To: "Philip J. Koenig" Cc: stable@FreeBSD.ORG Subject: RE: Hardware RAID vs vinum In-Reply-To: <20020529084340715.AAA489@empty1.ekahuna.com@pc02.ekahuna.com> Message-ID: <20020529022736.P472-100000@mail.allcaps.org> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-freebsd-stable@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.ORG Scott_Long@adaptec.com writes: > Greg, the companies that make RAID hardware are not filled with a > bunch of idiots. Please, people, no assumption of idiocy is required at all. Just add in some economics. Hardware isn't free. At the bottom end, let's choose something like the 3ware Escalade -- a $300 card which accesses 4 IDE drives. On at least some of the Escalade incarnations, there were 2 *big* XILINX FPGA's (moving that much data around requires a *lot* of pins) and an ASIC. When you add up the 3 chips, connectors, PCI card, cables, etc. there isn't a whole lot of money left over to make a profit (probably why 3ware actually thought about pulling the cards as a retail item). Consequently, you're going to try to make do with the memory you've got available in the FPGA's and ASIC as well as living within the available compute power. If it means limiting the stripe size and algorithms, so be it. You might think that some of the SCSI RAID cards at $600 or so can dodge this problem. However, if they use RAM and do battery-backup, all stripes in flight have to be backed up (remember, everybody likes the fact that SCSI can handle multiple transactions). The bigger the stripe size, the more RAM required to hold the same number of transactions. Otherwise, the card will start to block on transactions if it wants to back them all up. I can imagine that attempting to both battery back and move multi megabyte stripe sizes takes some serious compute power and memory. Probably beyond what someone is going to be able to put onto a $600-$800 card and make a profit. Now, I would be more interested in what someone like NetApp who have fewer economic constraints and some custom VLSI actually do from the stripe size perspective. However, I don't see anything in their product literature. -a To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message