Slashdot Mirror
The Ultimate Linux Box 2001
savaget points to this Linux Journal article which covers building a superior personal computer for general usage. See if you agree with the choices that Rick Moen, Daryll Strauss and Eric Raymond made in building their dream box.
I realize the SCSI disks, especially the close to "SCSI 3" mentioned in the article, would decrease disk latency, but is it really that much different than 7200 or even 10000 rpm ATA100/ATA133 drives? An unless you have onboard SCSI, you have to go through the already busy PCI bus. As far as I'm concerned, it's not worth the price difference.
Okay, Raymond isn't a millionaire any more, either. But he does have corporate backing, which is a hell of a lot more than I've got. When I feel like dropping 15 large on a personal computer, I think I'll go for an OS a bit more upscale than Linux. Solaris, maybe.
Anyway, "dream" is the key word in the title of the article. No real Linux users (mostly college students, AFAIK) can afford a PC like ESR has designed. And I'm not sure what they'll accomplish by "dreaming" about the "ultimate" Linux box when the whole point of Linux is to be able use whatever old, junk hardware you can scrounge.
Am I missing something?
I know RAID is overkill for most workstations, but so is a DDS drive and seperate home and system drives. If you want fault tolerance, (the stated reason for two drives) having one system drive and one home drive with no RAID means you spend your money only to become twice as vulnerable to downtime due to drive failures.
If you want to avoid downtime, especially if money is no object, get a RAID controller and have a single filesystem mirrored over two physical disks. Not only will it be more reliable, it will be faster too.
"Today's flatscreens also have a really coarse dot pitch with sharp square pixels. As far as I'm concerned, that puts them out of the running for the ULB. I do a lot of writing and, not infrequently, my own typesetting; I want to be able to preview two pages of Postscript at actual size and have the fonts look good."
I'm sorry, but has this guy ever seen a high-end flat panel? I personally own an SGI 1600SW, and not only do you not see the pixels, but you can also preview two Postscript pages side-by-side with its 1600x1024 widescreen aspect ratio. Of course, SGI stopped selling it (*sigh*). But there are other excellent flat panels out there, like the Samsung line that lets you run a TV signal in and do picture-in-picture. I've seen the Samsung ones up close, and they have wonderful image quality. Apple also makes some excellent flat panels (does anyone know whether there is an adapter to run them on PCs yet?)
All I'm saying, is while there are still plenty of reasons to run CRT's, in a "cost-is-no-object" type of article, you should at least consider the high-end flat panels.
P.S. I've seen the dual 1600SW setup, and it is STILL, to this day, the only monitor setup that ever made me speechless with its absolute beauty.
Well designed controllers like the escalades provide out of order execution, scatter gather, etc at the controller level, and offer a fully switched bus for all data. The 7000 series also have 64bit PCI support (and actually utilize it).
Forget the HPT36x and 37x controllers, as well as most Promise controllers, all the smarts is in the driver software and they suck performance-wise. High-end Adapted controller appear to be ok, but they are pricey compared to the 3ware controllers last time I looked.
One controller with one or two drives may be faster with SCSI, but dollar for dollar, 3ware and IDE walk all over them (particularly with database servers where you want a few spindles to minimize blocking seek activity.
I see a mention of a 56K modem, but no mention of an ethernet card. Perhaps ESR didn't have enough money to get the "ultimate network connection" to go with his "ultimate Linux box"?
> the whole point of Linux is to be able use whatever old, junk hardware you can scrounge.
Maybe that's YOUR whole point in using Linux, but it sure as hell ain't mine! If that's the way you feel, you'd be better off getting some nice DOS 3.11 disks somewhere.
I'd rather build a nice, fast 3Ware Escalade RAID array with 4 or so fast, big, and cheap IDE drives... and still save the CPU cycles.
The point of going with SCSI optical drives has more to do with the nature of the bus than the measured or perceived throughput.
.. IDE vs. SCSI.
.. without any space consuming and time consuming mastering to hard drive.
> wouldn't the new plextor 24x be way faster even if it is ATAPI?
First of all, you can't realistically burn media at 24x. The current media technology doesn't support it. The 24x drives only READ at that incredible spin rate, so don't get stuck on the spin rate. The focus here is bus-type
To illustrate the point, with the buffering capability and the resulting sustained throughput of even a mediocre SCSI flavor, you can read directly from a SCSI ROM and write directly to a SCSI writeable (CD-R, RW, etc.)
The numbers are looking better for ATA, but it's still not there.
--
Allen Gray
I expected more from the big wigs.
/pixel) = 1718 pixels wide
The "resolution" of a CRT is given by (veiw size)/(dot pitch). Any more pixels than that is literally wasted because the screen can't resolve one from another. The Mitsubishi 21" CRT he suggests has a view area of 20.3 inches and a dot pitch of 0.24 mm which works out to
20.3 in * 4width/5diag *25.4 mm/in / (0.24 mm
And he suggests running this at 2048 pixels wide? Sure memory is cheap, but bus bandwidth is teeny on PCs. Display what your monitor will do and no more. Also if back off the resolution a bit you could bump it up to 85Hz.
On an LCD a pixel is a pixel, and they're sooooo crisp compared to a CRT. They say the pixels are blocky, the rest of us call that clarity. Awesome clarity compared to a CRT.
I hate crap like this because these guys are supposed to be authorities, but they're spoiled brats whose hardware visions are 5 years out of date. Sure I'd like to use SCSI for everything, but get real. Looked at HD prices lately?
And apparently these guys haven't used a Contour keyboard (don't have a link off hand). I've put my hands on one, and you meld wih these babys, no stretching for keys or shifting your hands around, it's just BAM!
My box (single Athlon 1Ghz, IBM 7200RPM IDE drive) compiles kernels in 5m 45s (make clean; time make dep bzImage on 2.4.13pre1). Pretty good in comparison, considering that this box cost me ~$AUS2000, or perhaps $US1000. I'm using a nvidia geforce card and the emu10k1 driver, and the machine is rock-solid. It would also be a lot cheaper now - high-speed Athlons have come down a lot in price.
I understand the point of having the *ultimate* (rather than just "good") machine, and I realise that kernel compile speed isn't the most wonderful of metrics, but it does drive home the point that the more you pay, the less of a performance advantage you get. There's a price / performance sweet spot, and it's certainly not at the ultra high end.
The only thing I'd add would be a DVD drive - perhaps another $AUS170 for a cheapie Pioneer IDE model.
I'm not sure wether I should laugh or cry.
Linux can definetly be used as a multimedia OS. The only thing it still can't do as well is edit videos (Don't start with Broadcast 2000) and professional audio editing.
Linux is definitely a multimedia OS, unlike BeOS, which I have to say, is dead.
Sigh. More sickeningly pompous prose from ESR in a completely useless article. Building a $7000 personal computer is a pointless exercise which almost nobody who reads Linux Journal could afford to do. All of his "insightful research" and "surprising discoveries" is stuff you could easily find out on Usenet or the various hardware enthusiast sites -- the last time I built a machine (well equipped Tbird 1.4 for under $1000, tyvfm), I researched all the bits and pieces before buying them and avoided all of the headaches he's complaining about. And he can't even build the stupid thing himself:
You could build the ULB yourself from scratch. But unless you're either a very experienced hardware hacker or seriously interested enough in having a learning experience to accept possibly trashing some expensive parts, maybe you shouldn't. I wouldn't.
Way to encourage the hacker ethic! Yeah! Let's all run out and pay someone to do stuff for us, because everyone knows work is hard. With hardware prices as low as they are, it's a perfect time for people to "hack" their own hardware and build a powerful machine on a budget even a college student can afford. That would make an interesting article, but this one is simply, to use a phrase ESR seems to enjoy, an exercise in mental masturbation.
I just created a page where I post comments about some hardware I used, telling if I am satisfied with it or not.
I suggest that you do the same, so we can all make better buying decisions.