Hal is a hardware engineer who sometimes programs. He is the former editor of DTACK Grounded and can be contacted through the DDJ offices.
The city of San Jose is going to dig a long ditch that will pass near my house, according to the map published in the local newspaper. The minimum feature design rules for this ditch will be about 2x106 microns (this technical detail should allow me to sneak a ditch past DDJ editors). Nothing unusual? Well, I live in Santa Clara, and it is the city of San Jose digging the ditch.
You see, us folks in California are short of water. Rice farmers in California aren't short of water--they get 85 percent of the state's ample supply. It takes three gallons of water to grow each grain of rice. The California legislature has more friends of rice farmers than friends of people, so there's not much water left over for people. That's why San Jose is going to dig the ditch.
The ditch is going to carry reclaimed waste water (from sewage plants) to be used exclusively for watering lawns, mostly industrial-park lawns. Given that rice is more important than people, this is a good idea.
But San Jose officials have a better idea: As long as they're digging a long ditch that meanders all over the place, even into the city of Santa Clara, why not run fiber-optic cable in that ditch? Yes, San Jose is going into the cable/networking business as an entrepreneur.
San Jose is cleverly going to tax anybody other than the city of San Jose who competes in San Jose using fiber cable. A fabulous idea: Tax your competitors! (Wait until Borland and Novell hear about this.) The grounds are that fiber cable is a utility, and utilities can be taxed by cities. This absolutely assures that my section of the ditch, located in Santa Clara, will be taxed by the city of Santa Clara. (There is no municipal idea so powerful as a new tax.)
Al Gore tells me that the Information Highway is coming. With video capability added, the Highway needs fiber-optic cable. It'll arrive complete with a lot of federal censors and regulators. Schools and libraries will be connected for free, but the feds won't foot the bill. Who'll pay for this largess and for all the accompanying bureaucracy? Looked in the mirror lately?
I turn on Charlie Rose on PBS and I hear some talking heads tell me that a consortium of Baby Bells are trying to recruit Howard Stringer, head of the CBS TV network, to form a new, interactive-video cable enterprise. This is typical of a dozen credible (meaning they can reasonably expect to raise a few billion dollars) proposals; there are several dozen less credible.
The well-known futurist George Gilder wrote "Telecosm: The Bandwidth Tidal Wave" in the December 1994 Forbes ASAP. It looks in part very much like a public-relations release for "Tiger," a video-on-demand system backed by Bill Gates, of whom you may have heard.
I'm afraid that some day soon I'll hear pounding on my front door; when I open it, I'll find 17 people with 17 different cables competing for my attention. The problem with this prospect is that they'll be thrusting 17 different invoices at me.
The thesis of Gilder's article is that network bandwidth is increasing much faster than the internal bandwidth in our PCs. An accompanying graphic shows the crossover in 1996, after which network bandwidth makes the PC look like dogmeat: "_(network) bandwidth will expand 5 to 100 times as fast as the rise of microprocessor speeds_a firehose of gigabits (billions of bits)."
Gates's Tiger is an all-software, PC-based approach to handling video-on-demand and all that bandwidth. But maybe an all-software approach won't work.
Consequently, Gilder introduces us to MicroUnity, "a flagrantly ambitious (Silicon Valley) startup." MicroUnity is funded by, you guessed it, Microsoft. MicroUnity plans to build not a microprocessor but a mediaprocessor, one that can handle "not less than 400 billion bits per second" while replacing special-purpose (video?) multimedia devices. This mediaprocessor will have to be "hundreds of times faster than a Pentium." Flagrantly ambitious indeed!
Naturally (?), the video stream going out over those fiber cables will be compressed using MPEG standards. For this to happen in real time, you need a supercomputer that can execute video operations 1000 times faster than the raw video bits (according to Gilder, but this sounds right). At the receiving end, code division multiple access (CDMA) needs nearly as much video processing power.
At this point, when I'm about to be buried in gee-whiz, Gilder admits that Intel's Andy Grove "does not believe this possible." Whew!
I've often wondered just what those 17 fiber-optic cable purveyors will offer me when they converge on my front porch. One answer is, apparently, interactive TV. This has been tried on a small scale. I've yet to read of the vast success of interactive TV. The only time I physically respond to TV (as in to cheer) is on the (currently rare) occasions when USC stomps some football opponent, as it did in one of the minor bowl games last season. Hey, when I attended USC's University College (night school), USC was a football powerhouse.
Okay, interactive TV won't work for the mass populace. Neither will its sibling, home shopping. You need lots of money to do much home shopping and lots of money is what the mass populace don't got.
Instantly available CNN-worldwide TV news (our invasion of Panama was on CNN five minutes after the invasion started) is already here on the non-fiber-optic cable we've had all along.
I hate to tell you this, but the smart money is on video-on-demand, which translates into movies-on-demand, which means you can tap into your choice of available movies. Your choice will start at your convenience, to a resolution of 15 minutes. "Available movies" means the ones currently stored on the network supplier's hard-disk drives. The Jazz Singer and Citizen Kane won't be included. Given politics, Debbie Does Dallas won't be available either, even though such movies are very popular as video rentals.
How much will all this cost? The best estimate for video-on-demand over fiber-optic cable that I've seen so far is that it'll cost subscribers four times as much as existing TV cable, about $120 a month. That's a lot of money for federally regulated car chases on demand!
According to Gilder, three MPEG-compressed movies can be stored on a 9-gigabyte Seagate Barracuda disk drive. Given a "farm" of several hundred disk drives at Network Central, that's lots of movies "on demand." Will the movie you want to see tonight be currently stored on hard disk? (We don't need real-time video compression to load movies onto those disk drives, so we can lose the supercomputer.)
IBM originally planned to sell a very few hundred thousand PCs. Meeting this objective would have been judged a success. But IBM is selling millions of PCs every year, and this is widely considered a failure. Let's keep this in mind.
Gilder's Telecosm seems based on three tenets:
MicroUnity, funded by Microsoft, promised in 1994 to deliver 10,000 set-top mediaprocessors in 1995. Don't laugh, it could happen. (Seriously, everyone involved regards this as a high-risk gamble.) Here's how they are proceeding:
The mediaprocessor (MP) will have a very wide data bus. Pentium's 64-bit bus will look narrow in comparison.
The MP will be built using design rules about five times smaller than the Pentium; almost a tenth of a micron. This allows 25 times as many transistors for a given die size, as compared to the Pentium.
The MP will take an existing industry trend--CPU operating voltages have dropped from 5 volts to as low as 3.1 volts--to its logical conclusion and use a 0.5-volt power supply.
The MP will use an idea (already proved in lower-integration devices) that I think is brilliant: a signal layer using air, not silicon oxide, for insulation. The dielectric constant is far lower, and the impedance is proportionally higher. Even the signal-propagation speed is faster (trust me on this stuff; this is my pidgin). Regrettably, dense CPUs need many signal layers, and this approach, using air-insulated gold wires, will only work at the topmost signal layer.
All the aforementioned ideas are fundamentally good; the devil is in the details. If it were practical to make CPUs using 0.1-micron design rules, 0.5-volt power, and hugely wide data buses, then somebody would be doing it right now. Intel, the leading microprocessor producer, is just now bringing online a 0.5--0.4 micron production facility at its new fab in Rio Rancho, just outside Albuquerque. Production at or near 0.1 micron? In 1995, that's "Fantasy Island" stuff: "Da plane, da plane!"
For you hardware types, there's the ground bounce and nonexistent noise margins associated with vastly wide data buses and a half-volt power supply (the signal swing is, at most, ±0.25 volts).
Now for the good news: A technology tour-de-force similar to the mediaprocessor was successfully pulled off once. The year was 1982 and the company was Hewlett-Packard. HP invested the then unheard-of sum of $90 million in the design of a 32-bit CPU using design rules that were then as radical as those of the mediaprocessor today; see Figure 1.
This processor was highly successful. It had an unexpectedly high production yield, or proportion of good chips per wafer. HP restricted its use to HP's proprietary minicomputers and so succeeded, in terms of IBM's original objective for the PC. Had HP offered the processor for general sale, it might have ultimately sold millions and thus have been regarded as a failure: Why spend $90 million to make your competitors' computers run faster?
How will various network services (the Info Highway, home shopping, interactive TV, video-on-demand) share the necessary infrastructure and income stream? If video-on-demand is killed off by rented digital laser disks, are the remaining services worth $120/month to the mass populace? I hate to mention this in a family magazine, but the Politically Incorrect fact that pornography is readily available at the rental store but not over federally regulated networks is a powerful economic force against video-on-demand.
I think this whole thing will turn out just like HDTV: Politics and economics will swamp the technology. And you have doubtless noticed that after a decade of hoopla, you still don't have HDTV.
Figure 1 Microprocessor feature size (mm) versus date of first silicon (dashed lines indicate radical technology jumps). If network bandwidth is really increasing up to 100 times faster than microprocessor performance, why not wait two to six weeks for network bandwidth to increase so that video compression need not be used? That eliminates both the high-performance video-compression processor required for real-time MPEG and the need to decompress at the receiving end.
Look: Video compression is an expensive technique used to conserve a scarce resource--bandwidth. If bandwidth is really increasing so swiftly, it ain't scarce. Who else sees a credibility problem here?
**********
Nine-gigabyte drives go for $4500 these days, so it costs $1500 to store an MPEG-compressed movie. If magnetic-disk-storage cost per megabyte continues to drop by a factor of 2 each year, storage costs will drop to $200 per movie in three years.
The problem is, in three years we'll be able to buy a CD-like, digital laser disk that'll store a full-length movie for $15. I assume you'll be able to rent one of these disks for $2 to $3, thus replacing low-quality, short-lived video tape with high-quality, long-lived digital laser disks. This will be powerful economic competition for video-on-demand.
Do you want to buy stock in a company that's going to invest hundreds of millions, perhaps billions, of nonrecoverable dollars in the infrastructure needed to deliver movies via fiber optics rather than by digital laser disk?
Copyright © 1995, Dr. Dobb's Journal