From owner-freebsd-chat Thu Nov 6 10:48:16 1997 Return-Path: Received: (from root@localhost) by hub.freebsd.org (8.8.7/8.8.7) id KAA04472 for chat-outgoing; Thu, 6 Nov 1997 10:48:16 -0800 (PST) (envelope-from owner-freebsd-chat) Received: from schenectady.netmonger.net (schenectady.netmonger.net [209.54.21.143]) by hub.freebsd.org (8.8.7/8.8.7) with ESMTP id KAA04442 for ; Thu, 6 Nov 1997 10:48:03 -0800 (PST) (envelope-from postmaster@schenectady.netmonger.net) Received: (from news@localhost) by schenectady.netmonger.net (8.8.5/8.8.5) id NAA15752 for freebsd-chat@freebsd.org; Thu, 6 Nov 1997 13:22:57 -0500 (EST) Received: from GATEWAY by schenectady.netmonger.net with netnews for freebsd-chat@freebsd.org (freebsd-chat@freebsd.org) To: freebsd-chat@freebsd.org Date: 6 Nov 1997 18:22:56 GMT From: chris@netmonger.net (Christopher Masto) Message-ID: <63t1u0$fbd$1@schenectady.netmonger.net> Organization: NetMonger Communications References: <63spq5$6j7$1@schenectady.netmonger.net>, <199711061656.DAA00276@word.smith.net.au> Subject: Re: hardware Sender: owner-freebsd-chat@freebsd.org X-Loop: FreeBSD.org Precedence: bulk On Fri, Nov 07, 1997 at 03:26:37AM +1030, Mike Smith wrote: > > > > Dunno, I tend to agree with the SCSI advice for a couple of reasons. > > Having a FreeBSD box at home (from 2.0 through 3.0) with IDE, and > > several SCSI-based machines at work, it seems to me that the IDE is > > slowing my system down quite a bit. > > "seems"? Have you ever bothered to benchmark this? Are you comparing > apples with apples? No, I haven't bothered to benchmark this, hence my use of words like "seems". I try not to come across as authoritative on something I'm just giving an opinion on based on personal experience. Have you bothered to benchmark this? > Perhaps I should have posted over my other signature, the one that says > "high-speed data acquisition and realtime instrument control"? I *do* > try to keep a reasonable eye on things that impact on storage. 8) Oh, I see. You're more qualified to post your opinion than I am. The fact is that I do a lot of work on machines with IDE and SCSI and I decided to share my experience. I hear this "IDE is just as good as SCSI if you have only one drive" argument constantly, and frankly, I don't believe it. I suspect it has a lot more to do with scattered small files than sustained performance. I just installed 3.0-CURRENT on a brand new Western Digital 3.1GB IDE drive at home, and was extremely disappointed with the time it took to untar the ports directory (which, as you probably know, consists of thousads of tiny files and directories). I regularly do this kind of thing at work on SCSI systems and the difference is night and day. I also run a news server, so I have quite a bit of experience with "thousands of tiny files". I can report that on the news machine (which has an IDE boot and /tmp drive), heavy access on the IDE drive causes a lot more CPU load than heavy SCSI access (Adaptec 3940UW). > > The drive seems to just run a lot > > more, particularly when doing something like a CVS checkout. IIRC, > > IDE was basically a coupling of the ISA bus to a hard drive, and > > doesn't have nearly as much brains in the drive or controller as SCSI. > > You don't RC. IDE moved the register set of the WD1003 onto the disk; > arguably a Bad Idea but see my response to Joerg for commentary on > progress lately. The "smarts" in an IDE and SCSI disk are prettymuch > comparable. Then that doesn't explain the difference, and my theory is wrong. I'll have to read the specs at some point. Looking at /usr/share/misc/scsi_modes gave me the impression that there's quite a bit of firmware on a SCSI device, allowing it to do clever buffering, better sorting, scatter/gather, and a bunch of other things I don't know much about. > > The other reason I think SCSI is a good idea is that you can get a > > SCSI tape or Jaz drive or something. > > You can get ATAPI devices like these too. I don't recommend that. 8) Exactly. > > I'm currently struggling with > > the decision to dig into the kernel and get my Exabyte Eagle TR-3 > > "floppy tape" working under FreeBSD, or to just go SCSI and "get a > > real tape drive". I wish I hadn't tried to save a few bucks in the > > first place - the money I've wasted in upgrading and replacing IDE > > drives would easily cover the cost difference of having gone with SCSI > > in the first place. > > Uh, "floppy tape" != IDE. Duh. The point is that without SCSI, tape options are seriously limited. I happen to have a floppy tape drive that does not work with FreeBSD (as most of them don't). My options are either to make it work (I've written a few kernel drivers, but I don't know whether I want to spend the time on this) or "upgrade" to SCSI, which will give me a lot more choices. > And if you're talking about IDE disks you must be buyin' the gold > plated ones. Nope, but I've bought three of them, plus the tape drive. My point is that it's so savings if one buys IDE because one is trying to save a few bucks, and later decided to go with SCSI after all. > > Your mileage may vary (and apparently does). > > Please note my original perspective; IDE is price-performance very > competitive with SCSI, and becoming more so. IDE disks have reached > the point where they can be *seriously* considered for new systems in > classes other than "very cheap". I still think that depends on how you measure performance. Most benchmarks probably would show them about even, and for typical workstation uses, I suppose it doesn't matter. But for a news server, heavily-used shell machine, or other uses where there's a lot of noncontiguous disk activity, I would be concerned. Maybe unjustifiably - I think I'll cook up a benchmark to quantitatively test my allegation. > People making impassioned arguments based on their experience "years > ago" are strongly urged to reevaluate this technology at their leisure. > No promises are made as to absolute results, merely that consideration > for these disks should *not* be summarily dismissed as is the vogue. My experience is not "years ago". My experience is literally yesterday. I'm not summarily dismissing anything, just discussing my personal experience with these things. Perhaps in the data collection field, you regularly write large files sequentially, and it makes perfect sense to use IDE. -- = Christopher Masto = chris@netmonger.net = http://www.netmonger.net/ = = NetMonger Communications = finger for PGP key = $19.95/mo unlimited access = = Director of Operations = (516) 221-6664 = mailto:info@netmonger.net =