WILL NETWORKS BE THE BOTTLENECK IN DIGITAL MEDIA?
After confronting 3DTV, Ethernet’s inventor thinks we should be worried

Bob Metcalfe invented the Ethernet computer networking scheme at Xerox in 1973 and founded 3Com Corp. in 1979. He is now a visiting fellow in the Computer Laboratory at the University of Cambridge, England.

We all learned long ago from personal computers just how interesting interactive software can be without networks, working mostly on small personal databases, and sometimes on large static ones. And we learned recently just how much more interesting pcs are when they are networked, augmenting interpersonal communication and giving shared access to large dynamic databases.

Delivering media. Now consider the sum of accelerating developments in digital media — video telephones, HDTV, computer graphics, virtual reality, digital imaging, ubiquitous computing, interactive books, wireless palmtops, video conferencing, electronic publishing, dynamic multimedia, 3Dtv — and ponder how these real-time and data-intensive technologies might be delivered. Digital media without networks may not be media at all.

You probably agree, but have you thought enough about the networks needed to support new media? As a networking specialist, I worry that unless you think more about networks now, they will turn out to be your bottlenecks later, just when digital media approach commercial reality.

Calling the question. I want to get you thinking about what kinds of networks digital media will need. Who will make the huge investments, under what standards, and when, to establish a networking infrastructure? Are current networks as much of a bottleneck as they seem? What applications will drive network development?

I will use my favorite computer network, Ethernet, and a proposed benchmark application, 3DTV — not to answer whether networks will be digital media bottlenecks, but to leave the question churning in your mind.

APPROPRIATE TECHNOLOGY

By “networks” do we mean broadcast television, cable TV, telephones, computer networks? Well, yes, all of them, especially as they are converging. Certainly all are going digital, but is this convergence enough?

Given the magnitude of investments in network infrastructure, and given the concern for standards that such investments imply, the convergence will be slow and, if you do not get involved today, the convergence may be toward networks inappropriate for digital media.

I am a computer networker, and so it is no surprise that I think computer networks are the closest of today’s network technologies to what we’ll need for the commercial reality of digital media. Digital media are being developed on computers, are they not? And computer networks are digital to the desktop. They are also “packetized,” switched, two-way and increasingly high-speed. So I will proceed assuming some extension of today’s computer networks will best serve digital media.

THE INFORMATION INFRASTRUCTURE

There are (in round numbers) 100 million computers on earth, about one for every 50 people. Most of these computers are personal and isolated, but suddenly 10 million of them have become connected through an internationally standardized high-speed networking technology called Ethernet.

By “suddenly,” I am speaking in geologic time and mean over the last 10 years, and with “high-speed,” I am bragging and mean 10 megabits — two books of ASCII text — per second (10 mbps).

FORERUNNERS FOR DIGITAL MEDIA.

The High Performance Computing and National Research and Education Network Act of 1991, signed into law by George Bush last December, aims to invest hundreds of millions of dollars in a high-speed digital backbone for interconnecting computer networks. These, I claim, are the forerunners of future networks for digital media.

This infrastructure of computer networks has been growing since 1969, the dawn of computer networking, when the U.S. Defense Advanced Research Projects Agency (ARPA) began development of computer packet switching technology in the ARPA Computer Network (ARPANET). ARPANET has evolved to become a rapidly growing worldwide computer network with more than two million users, many at workstations on high-speed local-area networks (LANs). These LANs are interconnected using lower-speed telephone circuits in a network of networks called the Internet.

Mostly outside the Internet, but increasingly connected to it, are well over a million personal computer LANs. The majority of these are Ethernets, to which 10 million PCs are connected. These Ethernets carry the bulk of traffic in today’s information infrastructure.

So, are today’s computer networks already a bottleneck?

NETWORKS ARE NOT TODAY’S BOTTLENECK

If you read the computer networking trade press, you will get the impression that today’s computer networks are a bottleneck. And you will be given the unconvincing assurance that much faster networks are just around the corner.

But beware. Most of what you read flows from network users who mistakenly attribute system speed problems to their networks. And the rest of it flows from network suppliers who find it convenient to use speed to sell higher-margin “future-proof” networking technology.

Watch the silence. However, in the vast majority of cases, today’s LANs, especially Ethernet at 10 mbps, are still nearly empty. The simple truth is that when network users sit idled, it’s because their file server disks are seeking. Or because their PCs are grinding through poorly implemented or incompatible network protocols. Or their operating systems are switching and swapping. If you doubt this, just hang a scope on the network cable of a bogged-down system, and watch the silence.

PEDDLING SPEED

In 1976, I had the pleasure of promoting the Ethernet LAN concept when the big argument in data communication markets was whether anyone needed modems at 1,200 bits per second. It was pointed out that computer users could not read terminal screens fast enough to keep up with even a 600-bps modem. Those selling 600 claimed 600 was enough, those with 1,200 said it wasn’t, and I was laughed off the stage for talking about networks more than a thousand times faster.

In 1981, I started selling Ethernet products against the LAN market leaders, then Arcnet from Datapoint and Omninet from Corvus. I was careful to push the fact that Ethernet was an official industry standard, but many buyers were happier knowing Ethernet was four times faster than Arcnet and 10 times faster than Omninet. Ethernet entered hypergrowth and soon Omninet and Arcnet fell off in popularity. All the while, as today, LANs everywhere were almost empty.

When is four faster than 16? Five years later, in 1986, when IBM began shipping its IBM Token-Ring LAN, I sympathized as IBM salespeople were forced into ludicrous proofs that their 4 mbps was somehow faster than Ethernet’s 10. Failing that, despite the fact that its 4-mbps rings were still empty, IBM was then forced to introduce its 16-mbps variant. And we all soon may be sorry to see IBM try for the third time to have a fast IBM Token-Ring, this time maybe at 64 mbps.

In 1992, the networking trade press is full of progress reports on Fiber Distributed Data Interchange (FDDI), an emerging computer industry LAN standard and Ethernet’s putative successor. FDDI is ten times faster than Ethernet at 100 mbps. Of course, FDDI has the same two problems that Ethernet had against Arcnet and Omninet: FDDI is still ten times more expensive than Ethernet, and most Ethernets are still empty.

BANDWIDTH TO BURN

FDDI does not impress Ian Leslie here at the Computer Laboratory of the University of Cambridge, England. Leslie and his students have shown me a closet into which Cambridge’s new campus network is routed. The closet contains 12 cables, each with three bundles of eight optical fibers. Since Cambridge can now run 500 mbps on each of these fibers, Ian’s closet can handle a total of 144 gigabits per second (gbps), or 14,000 books per second. I think of this as insurance for when our Ethernets fill with digital media.

It is easy for a genius to predict that Ethernet’s 10 mbps eventually will be too slow for the majority of network-based information systems. But only a fool will say with certainty in what year, or perhaps even in what decade, the development of computers, their peripherals and demanding digital media applications will make networks the bottleneck, as they were in the 1970s.

APPLICATIONS THAT FILL NETWORKS

The historical co-evolution of computers, peripherals, applications and networks has some interesting stories. One of them is about laser printing and the invention of LANs.

Fact is, Ethernet was invented to support laser printing. In 1973, at the Xerox Palo Alto Research Center, we felt that we needed Ethernet to run at disk speeds — a few million bits per second — to feed scanned-in, 500-dot-per-inch bitmaps to upcoming page-per-second laser printers. Simple arithmetic showed that 1,200-bps modems were not up to the task. And so the experimental Ethernet was attached to such a fire-breathing laser printer at 3 mbps in 1974. And in 1979, the speed of the standard Ethernet was upped to 10 mbps simply because it was not much cheaper to run any slower.

A bit of overkill. But nearly 20 years later, among today’s millions of laser printers, few run at a page per second, and fewer still at 500 dpi. Further, thanks to page description languages like PostScript, we do not often send scanned-in bitmaps to laser printers, but very much smaller computer-generated document files, with fonts removed. Maybe this is why most Ethernets, most LANs, are empty.

So what do we now think the network requirements are for digital media? This is a question that I am urging that you now think about. To get you started, let me go through one particularly challenging application, perhaps today’s equivalent of laser printing in 1973.

WHAT ABOUT 3DTV?

Stewart Lang here at the University of Cambridge is working on three-dimensional television (3DTV). He was already working on building displays for 3DTV, and now I have him worried about getting 3DTV images transmitted to his displays — I have latched on to 3DTV as a benchmark application for future computer networks.

As Lang started my tutorial on the subject, I was interested to learn that most of the “3D” pictures I see are not truly 3D, not even on all those new “3D” workstations we are hearing so much about. Most are two-dimensional projections of 3D objects. As you move your head or walk about, the projections don’t change — Mona Lisa’s eyes seem to follow you around.

Remember 3D movies? Two slightly different red and green views of the action are updated at 24 frames per second. Your eyes are 6.5 centimeters apart and combine the views in one audience-wide stereoscopic approximation of real life. But even these are not what Lang calls 3DTV.

No special glasses. Lang defines 3DTV as a screen that several people can view in daylight with no special glasses, and see behind objects on the screen stereoscopically when either the people or the objects move around. This is accomplished by giving your eyes different views depending on where they are.

Holograms present 3D by projecting an almost infinite number of views around a room. Lang tells me that a smaller number of views will suffice.

For a 3DTV screen in a small room, with a 120-degree maximum viewing angle, Lang calculates that 100 views will be adequate. So, he asks, what kind of network have you got that can carry 100 TV channels, one for each view, to a 3DTV? I ask, how many bits are we talking about here? He says that this is a number increasing with time.

TV IS A MOVING TARGET

Today in the U.S., normal NTSC television is acceptable. This is 644 x 483 picture elements (pixels) per frame, updated 30 frames per second. Lang estimates 18 bits per NTSC pixel, so I take out my calculator and get 644 x 483 x 30 x 18 x 100 to get a bit rate of … 16.8 Gbps for NTSC 3DTV.

But of course NTSC is already a bit old-fashioned. Soon HDTV will be the norm and this multiplies out to 120 Gbps. After HDTV, people will soon be unsatisfied with image quality anything less than today’s 35mm films, which works out to 750 Gbps in 3DTV — five times Leslie’s whole closet.

Your first petabit. Here comes what may be your first real sentence above a terabit. How much storage is needed for a two-hour, 35mm 3DTV film? About 5.3856 times 10 to the 15th bits, or 5.3856 petabits. Excuse all the significant digits, but at these exponents every digit counts for a lot.

So I am thinking that if 3DTV takes off (and Lang is not claiming that it will), then maybe networks will indeed be the bottleneck.

MITIGATING FACTORS: THE THREE CS

Just as with laser printing, there may be some mitigating factors between these gross bit rate calculations and what actually gets sent.

Compression. None of the above numbers accounts for any image compression. Certainly there is much compression to be done. Current work with compressed broadcast television at 1.544 mbps (T1 in telephone talk) indicates there is another factor of 10,000 to play with.

Cameras. Lang may be planning to display 3DTV, and I may be planning to transmit it, but where are we going to get 3DTV images from? The building of 3DTV cameras at these resolutions may prove difficult — I won’t say impossible. So, maybe the cameras will be one of the bottlenecks.

Computers. Without cameras, 3DTV images may for a long time be computer generated. If so, then it is likely that a concise PostScript-like 3DTV image language will be transmitted instead of the daunting bit streams above.

So you see that 3DTV may turn out exactly like laser printing. A scary projected application drives a technology development, and then turns out later to be much less scary. These mitigations may give us networkers the time we need to keep up. Maybe networks will not be the bottleneck in digital media. But are you worried?

Bob Metcalfe

MINNESOTANS LIKE FUN, TOO

I have to take exception to Peter Dyson’s article “Nintendo Takes Aim” (see Vol. 1, No. 8, p. 17). The tone of the article seems to suggest that because Minnesota residents are against using Nintendo machines to play the lottery, they are somehow anti-fun.

Having lived in both California and Minnesota, I have seen two states victimized by the lottery. The lottery is sold to a state on the basis of state revenue and supplemental income to a state’s educational programs. In reality, millions of dollars are being sucked out of the state economy and sent to New Jersey, home of the company that runs the games. In addition, California immediately cut the education budget by the amount of the lottery supplements.

Luckily, Minnesota has not adopted the same “anything goes” attitudes about funding for our schools that California has. It costs the state even more to rehabilitate the 5 percent of gamblers who become addicted. Soon you can purchase a lottery chance over a 900 or a 976 number. This is an inherently bad idea too, and it has absolutely nothing to do with Nintendo.

The bottom line is that nobody wins with the lottery. I have not even begun to describe the social evils of the lottery, but I think all of us have seen people buying lottery tickets with food stamps (which is illegal) or people who look homeless spilling a pile of change on the counter to buy into a dream. This was not the result of some governmental slip, it happens to reflect the attitudes of the voters of Minnesota.

Bryan Menell
Exact Systems, Inc.
St. Paul, MN

I’m no fan of state lotteries myself. Nevertheless, the amount of money wagered each year indicates that gambling is popular with a substantial segment of the population. Making legal, state-run gambling inconvenient –which is what the opponents of Nintendo and 976 betting want to do — just creates a market opportunity for illegal private enterprises. This is the lesson of Prohibition.

– Peter Dyson