From owner-freebsd-hardware Tue Jan 4 9:27:34 2000 Delivered-To: freebsd-hardware@freebsd.org Received: from feral.com (feral.com [192.67.166.1]) by hub.freebsd.org (Postfix) with ESMTP id B44A41515F for ; Tue, 4 Jan 2000 09:27:31 -0800 (PST) (envelope-from mjacob@feral.com) Received: from beppo.feral.com (beppo [192.67.166.79]) by feral.com (8.9.3/8.9.3) with ESMTP id JAA09690; Tue, 4 Jan 2000 09:25:04 -0800 Date: Tue, 4 Jan 2000 09:27:15 -0800 (PST) From: Matthew Jacob Reply-To: mjacob@feral.com To: Mitch Collinsworth Cc: hardware@FreeBSD.ORG Subject: Re: differences between SCSI and EIDE [was: wanna buy an EIDE harddisk ... 5400 or 7200 for home use (noise)] In-Reply-To: <200001041610.LAA15549@benge.graphics.cornell.edu> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-freebsd-hardware@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org On Tue, 4 Jan 2000, Mitch Collinsworth wrote: > > >anyway, thank you all for responding and sheading light on my > >confusion. i'd always thought that scsi was the better way to > >go, either fro the 'comercial' environment or the ever more > >demanding 'home' environment. > > Well this is actually an interesting question. My salesman says the > HDAs are the same in SCSI and EIDE drives, so reliability-wise there > should be no difference. Yes, but unless we have stuff that turns on the bad block sparing, or otherwise does defect management, from a managability point of view there's a massive difference. > like to better understand what is being said here. Do SCSI device > drivers typically initiate multiple commands from separate processes to > the drive without waiting for the previous command to complete? In other Absolutely. The limitation here is probably the filesystem and load mix you're using- under heavy multiuser load I've seen all 256 tags used up. > words the drive logic has it's own queue management? And EIDE drives > require their device drivers to perform all queue management and only > initiate a command after the previous one has completed? > > Is the bottom line result of this that the SCSI drive has a much greater > chance of servicing multiple processes during a single media revolution > while the EIDE will frequently take multiple revolutions to service the > same queue? On the other hand, the newer bigger drives are getting able to basically consume most available bus bandwidth. If the numbers I've seen recently for drives being able to do ~24MB/s off the platter are indicative of things to come, then another reason for using SCSI (shared interleaved usage of an I/O bus) is going away because the limit is moving from the primary PCI bus to the seconday I/O bus, and if you can fit 4 ~20MB/s or better drives into a system (consuming most of the usable PCI bus bandwidth while you're at it) at a fraction of the cost for an Ultra2 LVD bus (which maxes out at 80MB/s), then indeed why bother with SCSI? This scenario changes slightly with Fibre Channel because the command processing time overhead that you can't get away from in parallel SCSI goes away as well as the tag limit so you can run a higher command load per spindle with Fibre Channel, but FC is definitely very expensive and fragile (from a programming point of view). It'll be interesting to see what Ultra3 brings to the party in all of this. At any rate, as long a most systems are single 33Mhz PCI bus systems, I'm rather annoyed to find that EIDE has snuck in and gotten good enough to be more than just your dopey local root disk. -matt To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hardware" in the body of the message