Snap, Crunch or GigaPOP?
Online traffic jams spur cries of
doom--and plans for Internet II
The Internet, warn some eminences grises, is staggering
chaotically toward massive outages, perhaps even a total collapse.
Nonsense, retort others: the future has never looked brighter for
the global network. Both sides are correct. True, the explosive growth of the
World Wide Web is pushing Internet standards and switches near their
breaking points, while floods of information regularly back up the
network plumbing.
But for more than a decade, congestion has hung over the Net like
the sword over Damocles, poised to sever its connections. Last-minute
additions of more and bigger pipes have always averted crisis. This
time, however, the problems run deeper, and although technical
solutions are in hand, they will exact a price--and not just in the
figurative sense. The resulting economic tremors may well topple some
of the Web's shakier business plans, but they should also reshape the
Internet into a more efficient and reliable medium.
The source of doomsayers' angst is the Net's geometric growth:
by most measurements, it doubles in size every nine months or so. Such
rapid expansion creates three major threats to the system. The first
jeopardizes its ability to connect any two computers on the network.
The Internet does so in much the same way as an automated postal
system: computers wrap data into packages, stamp the packets with
addresses and hand them to automated postal clerks (called routers) to
deliver.
But the Internet's numerical address system has nothing to do with
location. The Net equivalent of 10 Main St. may be in Maine, whereas
11 Main St. is in Ohio. So each automated clerk has to look up
delivery instructions in a table for every packet it handles. Because
packets often pass through 10 or more routers before reaching their
destination, the time spent poring over large tables can jam up
traffic considerably. More alarming, routers' tables are
growing twice as fast as their ability to search them. Within two
years, that could leave the Net's postmasters with just two unpleasant
options: either toss some packets into the trash or refuse to add new
addresses (especially those for competing network companies) to their
tables.
Two recent innovations will postpone that Faustian choice. The
first was a stopgap measure: the agency that hands out Net addresses
has been pressuring network managers to organize addresses into
sensible groups--much like zip codes. That strategy bought enough time
to start using the second improvement, a scheme called tag switching, which was introduced in September by Cisco, the
company that built most of the routers on the Internet. Here the first
clerk to examine a package writes down explicit instructions for all
the other clerks that will handle it, saving them the time and trouble
of consulting their tables.
The second threat to the Net is that it may
run out of
numerical addresses altogether, bringing its geometric growth to a
crashing halt. Although the current addressing format theoretically
supports about 4.3 billion computers, large swaths of the numbers have
been given away but never used. By recycling old addresses and dipping
into reserves, the existing supply can probably be stretched into the
next decade--long enough to switch to new software, playfully named "Internet
Protocol, the Next Generation." IPng will allow every human on the
planet to have something like 100 network devices. That should suffice
for a while.
The final danger to the stability of the burgeoning Internet is
that congestion will slow data to a crawl, ruining plans for
fancy interactive games, cheap long-distance calls and grainy video on
demand. Because bottlenecks often occur at the switches deep inside
the Internet cloud rather than at the periphery where workers and
consumers connect, the problem will only grow worse as more people buy
PCs and fast modems. Slick routing tricks such as tag switching will
help for a time. And many of the companies who own parts of the
Internet's backbone are scrambling to expand it;
MCI tripled the capacity of its
segment this past summer. But demand will outstrip supply as long as
Internet access remains so inexpensive; MCI has also seen the flow
over its network swell 56-fold in less than two years.
As Microsoft Network, America Online and Prodigy get ready to join
companies offering unbeatable, all-you-can-surf pricing, some schools
and corporations with high hopes for the Internet are preparing to jump ship. In October a
group of universities announced plans to build Internet II, a
high-speed national network linking perhaps 50 research institutions.
The private network would connect to the Internet at "GigaPOPs"
scattered throughout the country. (A POP, or point of presence, is the
Internet equivalent of a post office.) But it would close its gates to
outside users in order to preserve enough bandwidth to work on
high-tech projects--such as telemedicine, distance learning,
scientific visualization and broadcast video of undergrads' dorm
parties--without the hassle of Internet congestion. Companies such as
Chrysler are rumored to be
toying with similar options to link factories with dealers and
material suppliers. (Private "intranets" exist, but they generally
link their far-flung locations using the Internet and are thus at the
mercy of Net-wide congestion.)
Internet II will not ease the pressure on Internet I directly by
more than a few percent, but it may have a lasting indirect influence.
University officials involved say they want to try out new pricing
policies and special delivery software designed to help guarantee
rapid responses and clear channels to those willing to pay for them.
These good ideas have been around for years. One, called resource
reservation protocol, or RSVP, is even
scheduled to appear this fall in Cisco routers and Intel videoconferencing
software. The hang-up has been billing: if the urgent data are
delivered partly by MCI, partly by Sprint and partly by Pacific Bell, all three need to
agree on systems to split the fees. Internet II, because it would have
just one backbone and one bill to pay, could test whether RSVP and
other priority schemes work at large scales, while punting on the
billing issue.
In the meantime, some networking companies, chafing at the thin
margins of their commodity business, will soon start offering
higher-quality Internet access for higher prices. No one knows how the
market will react when, inevitably, basic services slow as premium
customers are ushered to the head of the queue. If, as some insiders
predict, the companies that run the Internet's backbone soon begin
charging those on its limbs according to the amount of data they send
or receive, they will have little choice but to pass the costs along.
Forced to decide what is worth paying for, many customers will first
tune out images--thus destroying the fledging Internet advertising
business--and will then search more, browse less. Although this may
rob the Net of much of its charm, it would almost certainly prod it
toward greater utility.
--W. Wayt Gibbs, staff writer, in San Francisco
RELATED LINKS:
Imminent Death of
the Internet?
Internet
Engineering Task Force
Internet
Reorganization Scheme
Alternative to Tag
Switching
Internet
essays by Bob Metcalfe, inventor of Ethernet
|