The Poweredge project is currently held up for want of screws. Specifically, twenty-four #6-32 x 1/4” flat-head machine screws.

They’re needed to mount the SCSI drives into the hotswap trays (procured on eBay for a few dollars each); standard hard-drive mounting screws — which are almost always #6-32 pan-head — won’t work. The drive trays have the holes for the drive-mounting screws countersunk into the plastic sides of the trays, since they have to fit absolutely flush. (Pan heads will hold the drives into the trays, but the protruding heads prevent the tray from sliding into the hotswap bay, as I found out to my chagrin.)

A handful of machine screws ought to be an easy hardware-store purchase, but unfortunately, finding a really good hardware store — the kind of place with drawers upon drawers of nuts, bolts, and other small parts, as opposed to the more common “home improvement” store — is right up there with finding a good typewriter repairman. They exist, but they’re few and far between.

After making some phone calls, I found a winner in Fischer Hardware of Springfield, VA. When I called to ask about screws, they cheerfully informed me that not only did they have 6-32 x 1/4” screws (spoken with a tone that seemed to imply “of course, dummy, we have #6-32 machine screws…”, truly music to my ears), they had them in my choice of stainless steel, brass, or zinc, in both Phillips or flat drive, how many did I want of each? Now that is the sign of a decent hardware store.

So tomorrow I’ll drive over there and see about picking up a couple dozen, and then I think I’ll finally be ready to boot the beast up.

The 2300 is an interesting (and large, and heavy) beast. It’s all SCSI — no IDE here — and has both an onboard U160 channel and the option to add a hardware PCI RAID controller. Mine has that option (called the “PERC 2/SC”) installed, and connected to a 6-bay front-loading hotswap backplane for SCA2 drives. Unfortunately, all the drives had been pulled, along with their sleds, when I bought it. Bummer. (I understand not leaving the drives in a surplused machine, but really, taking the sleds? That’s a bit low.)

A quick peek inside showed that it was full of RAM — exactly how much I couldn’t determine, since the chips didn’t specify and the part numbers didn’t bring up any useful information when Googled — and had a single 550MHz PIII processor installed.

Since the machine has two slots, my first search was for an extra PIII processor to fill it out. eBay quickly came to the rescue; for less than a measly $5 (and that’s with shipping), I had a second processor.

A little more Googling turned up some good deals on SCA2 U160 hard drives; unfortunately not as inexpensive on a per-MB basis as modern ATA disks, but dirt cheap compared to what they went for only a few years ago. I opted for 4 73GB 10k RPM Seagates to start with, enough to set up a decent RAID-5 array, while still leaving some room for additional expansion later.

On the OS front, I’m still not sure whether I want to go with BSD — probably OpenBSD, since I have an official CD set, bought mostly on impulse a while back — or Linux. I’m more comfortable in general with Linux, and I feel like I’ll be able to do more with the server if it’s running Linux, but I’ve been looking for an excuse to delve more into BSD and can’t decide if this is when I should take the plunge or not.

I was playing around earlier today, trying to find the slickest one-line command that would back up my home directory on one server to a tarball on another. (If I didn’t care about making the result a tarball, rsync would be the obvious choice.) I started to wonder whether it was possible to run tar on the local machine, but pipe its output via SSH to a remote machine, so the output file would be written there.

As is so often the case with anything Unix-related, yes, it can be done, and yes, somebody’s already figured out how to do it. The command given there is designed to copy a whole directory from one place to another, decompressing it on the receiving end (not a bad way to copy a directory if you don’t have access to rsync):

tar -zcf - . | ssh name@host "tar -zvxf - -C <destination directory>"

Alternately, if you want to do the compression with SSH instead of tar, or if you have ‘ssh’ aliased to ‘ssh -C’ to enable compression by default:

tar -cf - . | ssh -C name@host "tar -vxf - -C <dest dir>"

But in my case I didn’t want the directory to be re-inflated at the remote end. I just wanted the tarball to be written to disk. So instead, I just used:

 tar -zcf - . | ssh name@host "cat > outfile.tgz"

There are probably a hundred other ways to do this (e.g. various netcat hacks), but this way seemed simple, secure, and effective. Moreover, it’s a good example of SSH’s usefulness beyond simply being a glorified Telnet replacement for secure remote interactive sessions.

When it comes to geeky stuff, at heart I’m a hardware guy. I’m reasonably proficient at software configuration, and I can bang out a shell script or a little Python if there’s a need, but hardware has always struck me as more intuitive. Had I been born a bit before I was, I’d probably have become more interested in cars rather than computers, but sadly modern cars are fairly difficult to work on. Plus, mass production and Moore’s law, together with the ‘upgrade treadmill’ perpetuated by hardware and software vendors, have conspired to create an enormous, basically everlasting supply of IT junk, just waiting to be messed with and put to good use. As cheap hobbies go, as long as you restrict yourself to nothing that’s less than 4 or 5 years old, it’s about one step up from ‘trash art.’

So it was with that in mind that I found myself at a seedy self-storage facility last week, loading my latest acquisition into the back of my car. Via a corporate-surplus website, I’d picked up an old Dell Poweredge 2300 server for next to nothing. (Arguably, anything more than free is too much, but I was willing to pay a little to get one that was known to work.)

Over the next few weeks I’ll be playing around with it, with the eventual goal of setting up either BSD or Linux on it, and putting it to some sort of productive use (probably a backup server, if I can get the RAID system working) in my home LAN. Since information on the 2300 seems to be fairly limited, and there also seem to be a lot of them turning up on the used/surplus/come-get-it-on-the-curb market, I’ll periodically make updates with anything interesting I’ve found, and general progress.

[Prompted by this MetaFilter discussion.]

New technologies create new ways of communicating, thinking, and producing, but they also inevitably create new ways for con-men and hucksters to make an easy buck. Email brought us instantaneous, nearly zero-cost global communication; it also brought us spam. Webpages and search engines brought us more information at our fingertips than ever before in history; it also brought us domain squatting, tasting, typosquatting, blog and link spam. It’s an iron law of human nature that wherever there is a way to take advantage of a system for profit, someone will do it.

Amazon.com is poised to make the so-called “long tail” of book publishing available to all of us, by allowing ‘print on demand’ publishers to list their books in Amazon’s online catalog, and then print the copies individually, whenever an order comes in. It’s an idea with a lot of promise: by eliminating overhead, PoD allows books on incredibly niche subjects — which traditionally would have had a single short-run printing and then gone out of print, or not been printed at all — to stay available and in print.

But now this technology has found its own problem, eerily reminiscent of email’s spam and the web’s ad-ridden pages: automatically-produced ‘books’ consisting of database dumps on a particular subject. Like typosquatters who buy up thousands of domain names, knowing that it only takes a few ad hits to recoup the cost, or an email spammer who sends out billions of messages knowing only a few will lead to sales, a ‘titlesquatter’ can create thousands of ‘books’ in a database like Amazon’s, each on an almost ridiculously-niche subject. If an order comes in, the information is quickly assembled from publicly-available sources and the tome is sent out.

Phillip M. Parker, a professor of marketing at INSEAD, seems to be taking this route. He has over 80,000 books listed on Amazon, on subjects ranging from obscure medical conditions to toilet-bowl brushes. According to a Guardian article, they are written by a computer, at a rate of approximately 1 every 20 minutes.

Although some of the books do get positive reviews (not that this is saying much; Amazon’s review system is anything but unbiased), even the books’ supporters note that they are mainly compendia of Internet sources. This review, on “The Official Patient’s Sourcebook on Interstitial Cystitis” which retails for $24.95, is fairly representative:

I was very disappointed when I reviewed this book. It was almost as if the author(s) went to a search engine, and the NIH’s Medline, and the National Library of Medicine (PubMed) did a search for IC then made a book out of the results. … In my opinion, just a few hours on the web “today” will yield more current and useful information than that provided by this book. For those seeking information on IC, I suggest a search on “google.com” instead.

Others are more blunt:

The is downloaded copy of the NIAM website, and a list other research websites. I learned more from Google.

Although there may be a place and a market for ‘sourcebooks’ of this type, when they are clearly described and marked as being machine-written or -compiled, judging from the reviews it seems as though many consumers are purchasing them expecting more, and are consequently disappointed. This is bad news for print-on-demand, and the ‘long tail’ in general: if Amazon and others do not work to keep the content of their catalogs high, consumers may learn to mistrust anything that’s not highly ranked in sales numbers. PoD already has a poor reputation within the publishing industry, and if machine-generated books with plausible-sounding titles become more common, to the point where users have to sort through dozens of infodump ‘sourcebooks’ to find one offering new information, the situation could get far worse. At worst, it could turn users away from reference books completely — why bother buying reference books, if the majority of them just reprint what you can find in an online search anyway?

Although nothing that Parker is doing is illegal or even contrary to Amazon’s current policies, it makes sense for Amazon and other retailers that catalog PoD books to nip this behavior in the bud, before it becomes a full-fledged epidemic. If there’s anything that we should have learned from email and web spam, it’s that what begins as an oddity and an annoyance can quickly become a major waste of time and resources.

Conservative political strategist and blogger Patrick Ruffini has an interesting insider’s take on the fatal flaw of the Romney strategy. It was written on February 2nd, and seems even more relevant now — with Super Tuesday in the rear-view mirror — than it did then.

Huckabee and McCain represent two very distinct sides of both the Republican party and the ‘conservative’ movement in general. Huckabee is traditional and appeals to the base; McCain appeals to moderates and fence-sitters. That they are fundamentally different candidates is well-understood; this has basically been the nature of the Republican party since 1980 or so, and candidates’ overall success has basically been measured by how well they reconcile these two groups.

Enter Mitt Romney: onetime moderate, blue-state governor, Yankee Republican, entrepreneur. Realizing perhaps that it would be impossible for him to ‘out-liberal’ McCain without opening himself to accusations of being the Republican answer to Joe Lieberman, he made the strategic choice to place himself to the right of McCain and compete instead for the social conservative vote.

I thought and continue to think that this is a move requiring a whole lot of cojones. I’m not sure it was a good move, but you have to at least appreciate the inherent audacity. In theory, it’s pretty brilliant, but as good old Carl von Clausewitz once said, “Theory becomes infinitely more difficult as soon as it touches the realm of moral values.”

McCain is the Coca-Cola of GOP candidates, always performing at a consistent 30-40% … McCain does well in swing counties and liberal-leaning metro areas, but surprisingly, he doesn’t tank in rural, Evangelical areas. But Romney does.

My suspicion right now is that history will remember Romney’s bid as an interesting, but ultimately unsuccessful, gamble. What he probably could have been best at — wooing moderate voters and staking out a reasonable plank on both social and fiscal issues, backed with lots of past performance — was crushed as McCain and Obama both moved towards the center from opposite directions.

EDITED TO ADD: Romney dropped out earlier this afternoon, but has currently not pledged his delegates to any other candidate.

If you’re a Mac user, even an occasional one, and have been waiting with bated breath for a version of TrueCrypt, the wait is now over. (Okay technically it was over on 2/5, two days ago.)

TrueCrypt 5.0 includes Mac OS X native versions for both Tiger and Leopard on both PPC and Intel architectures, and the files it produces are binary-compatible with TrueCrypt for Linux and Windows.

Its use is not quite as straightforward as Apple’s Disk Utility, but in return it offers a far greater array of features, plus compatibility that Apple’s proprietary encrypted .dmg format lacks.

One of the most widely-touted features is the ability to create invisible ‘hidden volumes’ within the free space of other encrypted volumes. Another is the choice of ciphers; while Apple supports AES, TrueCrypt offers AES, Serpent, Twofish, and combinations thereof (any two or all three at once, operating in serial on the same blocks). It also allows a choice of three hash algorithms, including the openly-developed RIPEMD-160.

Just for test purposes, I created a 4GB volume using Twofish and RIPEMD-160; actual volume creation on a Dual 2GHz PPC G5 ran at about 9.5MB/s. Copying to it seemed to be around 3MB/s on average, with excursions up to around 5-6MB and periodic short stalls. Overall, a 600MB file took about 4 minutes to move onto an encrypted volume.

One of the only missing features in the Mac version is the inability to create sparse files that expand in size as they are filled. (This is possible with .dmg files although it requires the use of the command-line ‘hdiutil’ to do it.) I’m not clear on the details but it sounds like TrueCrypt’s sparse file support relies on the NTFS filesystem. But given the problems I’ve had with sparse files in the past (they get easily mangled when copying across filesystems and various OSes), and the low cost of storage, I’m pretty content sticking with static files.

Overall, this is a big win both for Mac users and TrueCrypt users in general, since it makes the product that much more flexible overall. As an encrypted container format I think TrueCrypt is fast becoming the de facto standard, and now you can put a FAT-formatted .tc file on a USB stick and be pretty much assured that it will be readable no matter where you go.

I’ve wondered for a while if it’s possible to construct a swappable external hard drive by putting one of those cheap removable IDE drive drawers inside a 5.25” external FireWire enclosure, but not enough to actually go out and buy all the parts and then have to return them when it didn’t work. However, thanks to the wonder of the Internet I was recently able to pick up both pieces for under $20 from the clearance section of Geeks.com.

Short answer: it actually works. (Even spindown.) For twenty bucks plus shipping and a spare IDE drive, you can make yourself a functional analog of a Quantum GoVault, suitable for all sorts of disk-based backup tasks.

In retrospect, I’m not sure why I thought it wouldn’t work. I think I was assuming far more complexity on the part of the IDE drive drawer than actually exists; I figured they were an actual backplane with some logic to let you hot-swap the drive. In reality they’re nothing more than a big Centronics connector for the power and IDE connections, a few LEDs, a fan, and a key-operated on/off switch to keep you from ripping the drive out while it’s spun up. It doesn’t get much more basic than that. Once you have the drive inserted and locked, the FireWire bridge is none the wiser.

Where things get messy is if you actually try to hot-swap the drive while it’s in use. Because the FW bridge isn’t designed for hot swapping on the IDE side, bad things tend to happen when you remove the drive and reinsert it while the FW bridge is on. However, as long as you power off the entire external enclosure (rather than just the drive), then swap drives, then power back on, everything works nicely. I think of it as sort of a ‘warm-swap.’

I prefer the drive-tray-and-enclosure solution to just buying multiple 3.5” enclosures because the 5.25” one — unfortunately no longer available from Geeks.com but easily found elsewhere under its model number, “PM-525F2-MOS” — has significantly better heat characteristics than most cheap 3.5”s (it’s aluminum and has a fan, for starters; the unvented plastic ones eat drives for breakfast), and the cost per additional drive is much lower.