Dr. Dobb's Journal January, 2005
The big news recently was the ANSARI X PRIZE (http://www.xprize.org/) was won by SpaceShipOne. Now I have no doubt that all but a few DDJ readers find this interesting and exciting, it may be less so obvious why it's particularly important to the computer community.
First, of course, it was paid for by one of our own—Paul Allen (and built by Burt Rutan). When the computer revolution began, IBM's vision of the future was 10-to-100,000 big central computers, all operated by white-coated technicians, serviced by IBM Customer Engineers, and running software leased from IBM or an authorized IBM partner. Bill Gates and Paul Allen had a different vision: A computer on every desk, and in every home, and in every classroom. While it is likely that Moore's Law would see to it that the great computer revolution took place without Microsoft, it's pretty certain that it wouldn't have happened so quickly without that vision.
We forget that a decade after the introduction of the PC there were serious debates about how much education it took to become computer literate, and computer literacy instruction techniques were taught in our colleges of education. UNIX gurus battled over the purity of programming, there were language wars, and many thought that while computers in homes and offices were useful, they wouldn't become ubiquitous because they weren't user friendly. That battle has long been settled, but it wasn't obvious to all that things would go as they have.
Even before they became ubiquitous, small computers had their effect in bringing down totalitarian states. Arthur Koestler once predicted that a sufficient condition for the dismantling of totalitarianism would be the free exchange of ideas within a totalitarian state, and the advent of the small computer made it certain that computer users would be able to freely communicate. The Falkland Islands War made it obvious to everyone that economic and military power depended on using small computers, while western grand strategy put more and more pressure on the USSR to modernize its weapons at great economic costs. The choice was clear: Adopt computer technology and let your intelligentsia communicate freely, or forfeit all pretense of military and economic power.
The USSR accordingly collapsed, and China changed irrevocably into something other than what it had been (although staying uniquely Chinese). It was not, as Francis Fukuyama rather precipitously proclaimed, the end of history, but the totalitarian state built around total control of information flow did slide into the dustbin of history and is unlikely to emerge short of a new Dark Age. The computer revolution accomplished at least that much.
Some years ago Esther Dyson's keynote at a meeting of the American Association for the Advancement of Science proclaimed the end of "the official story." Information monopolies don't work. If the story is false or leaves out a lot, the corrections will appear, and quickly. We have recently seen many instances of that.
The World Wide Web has stitched the entire civilized world into one enormous information pool; and the Web itself is evolving techniques for sorting information from misinformation. That process isn't anything like perfect, but it is in development.
SpaceShipOne was paid for by money Paul Allen earned from the computer revolution, but that was hardly its main debt to the little beasts. The ship was designed with CAD programs running on off-the-shelf PCs. More importantly, though, all the flight simulation was done by a bank of networked off-the-shelf desktop Personal Computers, mostly from Dell.
SpaceShipOne made only six powered flights. Each one went higher and faster than the previous flight, and the last three reached space, the first commercial ship ever to do so.
While the ship itself flew only six times, the pilots had made hundreds of flights using highly realistic flight simulator programs. It was the only way they could learn to fly the ship.
In my early days in Projects Mercury, Gemini, and Apollo, flight simulators were expensive. Indeed, airline flight simulators cost almost as much per hour to operate as actual airliners. Computers were expensive to buy and expensive to operate. Now nearly any reader can afford a SpaceShipOne flight simulator. The actual software contains some proprietary information that Vulcan Ventures (the closely held company through which Paul Allen financed this effort) hasn't released, but the actual software isn't complicated and was built around standard programs. It runs on standard Windows machines using commercial video cards, and costs no more to operate than any other PCs.
The flight simulator was important for another reason.
SpaceShipOne's flights into space weren't all smooth. The first flight to space on June 21, 2004, began with a severe yaw anomaly and experienced some damage to nonstructural components. That was simulated and analyzed in the flight simulators. The second flight to space—first flight of the X PRIZE competition—was marked by a roll anomaly, with the ship rolling 29 times on its way up. That too was analyzed and simulated. The third flight to space was flawless as it won the ANSARI X PRIZE.
It couldn't have happened without the simulators, and that kind of simulation wasn't affordable to small companies before the computer revolution.
Some years after Apollo successfully landed on the Moon I had an argument with the late John J. Pierce, then chief technologist for the Jet Propulsion Laboratories (http://books.nap.edu/html/biomems/jpierce.html). Pierce reminded us that Heinlein's original fictional path to space led through suborbital flights of reusable vehicles; this developed both institutions and infrastructures, ways of making money from suborbital and then orbital flight, communications and control networks, and so on. Only then could space travel develop.
At the time, I thought Pierce mad. After all, we had just gone to the Moon, and I was proud to have been part of Project Apollo. Now, 30 some years later, I see he was right. Apollo was a government funded stunt, and didn't lead us in the right direction. Perhaps it couldn't. Now, though, we have the computer revolution, and we are on the right path at last.
If you don't have gigabit Ethernet, you soon will—and it won't cost much. It's another of the fallout benefits of Moore's Law.
When we rebuilt Chaos Manor back in 1985 we designed the entire upstairs around computers. Of course, we got most of it wrong. For example, when we designed the new office suite, the best printer for a writer was still a machine that looked like an enormous typewriter and printed out manuscripts by impact: They came in several flavors, but in every case they printed on fanfold paper and worked at about 1200 characters a minute.
The advantage over the more usual dot-matrix printer was that despite the ragged edges of the fanfold paper, the manuscript looked like a typewritten manuscript, and that was what publishers were used to and some of them demanded. Dot matrix looked like computer printing, and was generally harder to read.
It took all night to print out the manuscript of a novel. The noise was deafening. Robert Heinlein once had a glass cage built around all but the keyboard of his IBM Executive electric typewriter because "I didn't want to be in the same room with a machine gun." That wasn't an option with the Diablo, later NEC Spinwriter, that I relied on for printing, so we designed a small closed room to hold the printer. It was originally intended to have a glass window in the door so you could see print operations in progress.
By the time construction was actually under way, Hewlett-Packard had come out with its LaserJet printers. They were so scarce there was a half year waiting list for them, but one of Dr. Bill Godbout's friends, Steve Rose, managed to order six of them, and Dr. Godbout talked him into letting me buy one. It came during construction, and was installed downstairs near my desk, and worked so silently that the "printer room" wasn't needed. We quickly modified that room to have its own small air conditioner, and wire channels to all the other parts of the house. It became the infamous "cable room" where we stored spare cables as well as where we kept the central server. Of course in those days of very early PCs, very early Macintosh, and the more robust and generally preferred S-100 bus machines, "server" had a different meaning.
So did networking. In 1985 the Ethernet chipset cost about $800, and that price looked stable. Bill Godbout and others in those days recommended ARCNET, because although it was slower and less technically elegant than Ethernet, it was a lot cheaper. The result was that I still have ARCNET coax pulled through various wire channel guides to the cable room. Needless to say, all that has been "abandoned in place." (Interestingly, ARCNET is still around, although it isn't all that popular. http://www.arcnet.com/.)
A few years later Artisoft and others brought the cost of Ethernet down from a few hundred to a few tens of dollars per system, and for a while we had Ethernet Thin-net coax cables, mostly strung overhead in what Roberta calls "computer decorator unfriendly" patterns. That worked at 10 megabits/second, faster than the old ARCNET. Moore's Law continued its inexorable pace, and 100 megabits/second Ethernet became available. It worked only on Category 5 twisted pair cables, but unlike Thin-net it would network in many different configurations. Those were the days of Ethernet "concentrators" (also called "hubs") that had both Thin-Net coax and twisted pair RJ-45 connectors. Over time those became switches, able to connect to both 10 and 100 megabit/second Ethernet connector cards. About then I had some of Alex's professional associates come in and pull Category 5 twisted pair Ethernet cable through much of Chaos Manor. That has served ever since.
Ethernet at 1 gigabit/second has been around a while. My last few Intel motherboards have gigabit Ethernet, but I paid little attention until one day I noticed a sale at Fry's: AirLink 5-port 10/100/1000/megabit Ethernet switches for about $25.00 on sale, with an 8-port for about $50.00. I had no idea how well those would work with my Category 5 cable system, but it would be cheap enough to find out, so I bought two 5-port and one 8-port switch.
When I got them home I opened the 8-port switch first, and discovered that it had to sit on a shelf: There were no little mounting holes on the back side of the case, as you will find on switches from Cisco, Hawking, and NetGear. Since the place I needed the gigabit switch required that I hang it on a wall, this was dismaying, and I left off doing anything with it until I could figure out where to put it. Then I got ready to go on a trip: One phase of this requires transferring a 1.6 gigabyte PST file to the laptop and sending another copy of that file to a backup storage machine. That takes about 11 minutes, and while my TabletPC doesn't have gigabit, the other systems do, and cutting the time to do either of those operations would be worthwhile.
When I opened the 5-port AirLink Ethernet Switch, I found it has mounting holes on the back. Only the 8-port lacks them. I quickly hung the 5-port on the bookcase where my other central concentrators and switches reside, and connected all the machines I thought capable of gigabit operations to it.
The AirLink switches have lights for each port to indicate what speed the device sees at the other end of its cable. I wasn't surprised to see that Lisabetta (the TabletPC) didn't have gigabit, but I was disappointed to see that Sable, an Intel D850EMVR2 dual-channel 3-GHz Northwood, was only 100 megabits/second. Anastasia, an Intel D875PBZ Northwood at 3.2 GHz, and Lassie, an Intel D865GBF Northwood, both had the gigabit lights, and later tests showed that all the Prescott systems have gigabit. Apparently my D850 was among the last to employ 100-megabit Ethernet chipsets.
When you go from 100 to 1000 megabits/second, you don't get 10 times better performance, of course. It's more like faster by a factor of three. Actual performance varies all over the place. With two machines connected through the same switch with no other tasks running in the background, the same 1.6 gigabyte Outlook.PST file transferred once in 263 seconds, and another time in 338, at 100 megabits/second. The same file took 135 seconds (first trial) and 71 seconds (second trial) transferring to a 1-gigabit machine. In neither case did CPU usage go above 25 percent, and the disks were large and nearly empty on both machines.
I have no explanation for the variance in times. Bob Thompson speculates that the Windows filesystem and TCP/IP stack have a lot to do with it. Anyway, it's clear that if you do a lot of big file transfers it's worth changing to gigabit switches if you already have gigabit capabilities in your machines.
Note that it is probably not worth the effort to replace your current Network Interface Card (NIC) with a gigabit card. Gigabit Ethernet is enough to saturate the PCI bus, so simply adding a Gigabit Ethernet PCI Card is less than optimum practice, particularly for servers (where you would really want Gigabit Ethernet).
Intel motherboards take the Gigabit Ethernet adapter off the PCI bus and put it on a CSA (http://www.buildorbuy.org/csa.html) or PCI-Express bus. The right way to upgrade to Gigabit Ethernet is to replace the motherboard. That may be a bit more than most of us want, although I probably will do that for my older servers.
In my case, I have four AirLink Gigabit Ethernet switches set up: two at various nexuses (nexii?) where my fastest machines connect to each other; another in the cable room; and a fourth out at the test stand where we set up new machines. The one in the cable room connects the others using the Category 5 cable we pulled several years ago.
Ethernet switches work this way: If two machines connected through the same switch want to talk to each other, the switch connects them without routing the signal anywhere else. If they are not both connected to the same switch, the switch sends the signals up the line to a switch that can find the system they're intended for. I have tested my gigabit connections both direct (two machines connected through the same switch) and indirectly (system to switch, switch to another switch then to a third switch, final switch to second system) and if there's any systematic difference in file transfer speed it's lost in the noise of other factors causing speed variance. Understand, I haven't been using any sensitive tools, just sending big files and recording the time it takes to transfer them.
My conclusion is that you'll get from 2.5 to 5 times the speed using Gigabit Ethernet (1000 Base T) rather than 100 megabit Ethernet (100 Base T), at least if you simply change to faster switches and you began with good quality Category 5 cable to begin with. Given that it's pretty cheap to do this, it was worth it for me. Note that Category 5 cable isn't rated for Gigabit Ethernet use, but as a practical matter it seems to be working; of course I have no wires longer than about 20 meters.
While I was at Fry's looking for something else, I decided to get some short Ethernet cables and thereby cut back on the rats' nests. I discovered that they sell 3-foot Category 5 cables for $4.99, and the same length Category 6 for $3.99. This seemed confusing because I would have thought Category 6 was "better" than Category 5; but no one at Fry's understood why they were priced that way, and for three cables I didn't really care about the price difference, so I bought Category 5 which I knew would work.
I should have bought Category 6, as you will find if you go to http://www.lanshack .com/cat5e-tutorial.asp where more is explained about Ethernet cables than most of us really want to know. Dan Spisak thinks there's a big push to get people using Category 6 cables and that's why the price break, but that's a guess. I doubt it made any difference, but in future, I'll get Category 6 cables, and if they speed things up I'll let you know.
The game of the month is a lament. I sure miss a lot of the old turn-based strategy games, particularly the battle simulations from SSI. The good news is that DOSBOX (http://dosbox.sourceforge.net/news.php?show_news=1) gets a number of those running properly. The bad news is that I gave away a lot of those SSI games that I wish I had kept, and I'll probably have to go haunt eBay to buy new copies now that I have ways to run them again.
The book of the month is Who Are We?: The Challenges to America's National Identity, by Samuel P. Huntington (Simon & Schuster, 2004). It's a deep inquiry into America's national identity crisis and ought to be required reading for anyone who is serious about self government.
The computer book of the month is David Pogue's MAC OS X: The Missing Manual (O'Reilly & Associates, 2003). If you want to connect your Mac to a Windows active directory system, you'll need this. Actually, you'll need this book for almost anything you want to do with a Mac; it really is the book that should have been in the box. If you have no interest in Macs, get Ken Milburn's Digital Photography Expert Techniques (O'Reilly & Associates, 2004), which is massively illustrated, readable, and full of information about taking better pictures with digital cameras. I only wish I had the time and patience to master this book; but just thumbing through it now and then helps.
DDJ