The coming broadband congestion
Web Informant #163, 8 August 1999
More and more of us are fortunate these days to have high-speed Internet access. I now enjoy Cablevisions cable modem at home and Wired Business/Covads DSL line in my office. Ive almost forgotten the sounds of a modem going through its dial-up routine, and my phone bills are no longer painful to pay. (Business users here in Bell Atlantic land have to pay for every minute of every ISDN call, local or not. That adds up when you connect to the Internet via ISDN, as I used to do from my office.)
But having a high-speed connection just moves the bottleneck for your access from your own immediate vicinity to some other place on the Internet. Overloaded servers and slow peering points will become more frequent as more and more of us obtain these connections. To show what we have in store for us, I turn to my friend Kim Maxwell (KimMaxwell@aol.com).
Kim and I serve on Covads advisory board, just so you know. And Kim has written a recent book about various high-speed networking technologies, applications, and markets, from which this essay is excerpted with permission. You might argue that since he wrote this piece two years ago, the number of Internet users has grown way beyond his original estimates of 45 million. However, his math and analysis still remain the same. Take it away, Kim.
What will happen to the Internet when millions of users try to access it at high speeds using cable modems and DSL lines? What will happen if thirty percent of Internet traffic becomes continuous rather than intermittent data? I will show you with some simple arithmetic a scary situation that could mean we are in for some very congested times ahead. Indeed, we may have to shift a majority of high-speed and other broadband traffic to local servers located a few
miles from these heavy users, to avoid the Internet altogether!
Lets do the math together. Let us accept the claim that there are 45 million Internet users today, but only 10 percent of them use the network at any one time. Their average access rate is 28.8 kbps and their network usage -- the percentage of time data is actually on the line during a connection -- is 25 percent. In a crude sense, the network must process, in aggregate, 32.4 billion bits per second to handle this load.
Let us then suppose that, five years from now, the Internet has 200 million users, but 30 million of them have residential broadband at an average speed of 1.5 Mbps. If we make the same usage assumptions, the Internet must process 1280 billion bits per second, of which 1130 billion bits per second goes to broadband users. In ratio terms, 88 percent of network bandwidth is now supporting 15 percent of network users.
To keep up with this traffic, the Internet must more than double every year, with an annual user increase over the same period of only 34 percent. But this picture is unreasonably hopeful. Residential broadband users are always connected, removing one obstacle to more simultaneous users, particularly during peak loading hours, which seem to be weekday evenings.
This may be counteracted by proportionally longer waits between screens, reducing network usage per user. However, always-connected users will increase signaling load compared to traffic, and higher speeds will promote more frequent file transfers, an application with 100 percent usage.
Furthermore, within five years more and more Internet traffic will be video and streaming audio. To make matters worse, cable modems and at least 50 percent of ADSL modems installed at that time will actually run at 6 Mbps, suggesting that average broadband speeds may be closer to 3 Mbps. If we stir in some of these new assumptions, the aggregate network demand approaches 6.3 terabits per second!
Under these assumptions 98 percent of network capacity is used by 15 percent of its users and the network must quadruple annually, to grow to 175 times its present capacity in just five years. To get some idea of how big this really is, the same network capacity running under todays access conditions of dial-up modems and the 10/25 ratios of concentration would support 7.5 billion users, more than presently inhabit the globe. But there will be only 200 million users, and only 30 million broadband users.
Ultimately, users pay the bills. We can say almost without qualification that the network cannot grow that fast; even if it could, no one will want to pay for it. Network designers will scoff at the simple-mindedness of this analysis, and make claims for economies of scale, reduced hardware costs, faster routing with IPv6, faster routers, and free fiber bandwidth. But bandwidth is not free and even Moores Law will not keep up with 200 percent per year growth.
Thanks Kim. You can buy his John Wiley and Sons book, Residential Broadband from Amazon at the location below: http://www.amazon.com/exec/obidos/ASIN/0471251658/davidstromswebin/
Next week Ill talk more about my experiences with using both cable and DSL access technologies, and where I see both going in the coming years. And, Ill give you my thoughts on which technologies are best able to handle this coming congestion problem.
Copyright © 1999, 2000 media.org.
Web Informant copyright 1999 by David Strom, Inc., reprinted by permission
Web Informant is ® registered trademark with the U.S. Patent and Trademark Office.
ISSN #1524-6353 registered with U.S. Library of Congress.