The heart of the Internet is a network of high-capacity optical
fibers that spans continents. But while optical signals transmit
information much more efficiently than electrical signals, they're
harder to control. The routers that direct traffic on the Internet
typically convert optical signals to electrical ones for processing,
then convert them back for transmission, a process that consumes time
and energy.
In recent years, however, a group of MIT researchers led by Vincent
Chan, the Joan and Irwin Jacobs Professor of Electrical Engineering and
Computer Science, has demonstrated a new way of organizing optical
networks that, in most cases, would eliminate this inefficient
conversion process. As a result, it could make the Internet 100 or even
1,000 times faster while actually reducing the amount of energy it
consumes.
One of the reasons that optical data transmission is so efficient is
that different wavelengths of light loaded with different information
can travel over the same fiber. But problems arise when optical signals
coming from different directions reach a router at the same time.
Converting them to electrical signals allows the router to store them in
memory until it can get to them. The wait may be a matter of
milliseconds, but there's no cost-effective way to hold an optical
signal still for even that short a time.
Chan's approach, called "
flow switching,"
solves this problem in a different way. Between locations that exchange
large volumes of data — say, Los Angeles and New York City — flow
switching would establish a dedicated path across the network. For
certain wavelengths of light, routers along that path would accept
signals coming in from only one direction and send them off in only one
direction. Since there's no possibility of signals arriving from
multiple directions, there's never a need to store them in memory.
Reaction time
To some extent, something like this already happens in today's
Internet. A large Web company like Facebook or Google, for instance,
might maintain huge banks of Web servers at a few different locations in
the United States. The servers might exchange so much data that the
company will simply lease a particular wavelength of light from one of
the telecommunications companies that maintains the country's
fiber-optic networks. Across a designated pathway, no other Internet
traffic can use that wavelength.
In this case, however, the allotment of bandwidth between the two
endpoints is fixed. If for some reason the company's servers aren't
exchanging much data, the bandwidth of the dedicated wavelength is being
wasted. If the servers are exchanging a lot of data, they might exceed
the capacity of the link.
In a flow-switching network, the allotment of bandwidth would change
constantly. As traffic between New York and Los Angeles increased, new,
dedicated wavelengths would be recruited to handle it; as the traffic
tailed off, the wavelengths would be relinquished. Chan and his
colleagues have developed network management protocols that can perform
these reallocations in a matter of seconds.
In a series of papers published over a span of 20 years — the latest
of which will be presented at the OptoElectronics and Communications
Conference in Japan next month — they've also performed mathematical
analyses of flow-switched networks' capacity and reported the results of
extensive computer simulations. They've even tried out their ideas on a
small experimental optical network that runs along the Eastern
Seaboard.
Their conclusion is that flow switching can easily increase the data
rates of optical networks 100-fold and possibly 1,000-fold, with further
improvements of the network management scheme. Their recent work has
focused on the power savings that flow switching offers: In most
applications of information technology, power can be traded for speed
and vice versa, but the researchers are trying to quantify that
relationship. Among other things, they've shown that even with a
100-fold increase in data rates, flow switching could still reduce the
Internet's power consumption.
Growing appetite
Ori Gerstel, a principal engineer at Cisco Systems, the largest
manufacturer of network routing equipment, says that several other
techniques for increasing the data rate of optical networks, with names
like burst switching and optical packet switching, have been proposed,
but that flow switching is "much more practical." The chief obstacle to
its adoption, he says, isn't technical but economic. Implementing Chan's
scheme would mean replacing existing Internet routers with new ones
that don't have to convert optical signals to electrical signals. But,
Gerstel says, it's not clear that there's currently enough demand for a
faster Internet to warrant that expense. "Flow switching works fairly
well for fairly large demand — if you have users who need a lot of
bandwidth and want low delay through the network," Gerstel says. "But
most customers are not in that niche today."
But Chan points to the explosion of the popularity of both Internet
video and high-definition television in recent years. If those two
trends converge — if people begin hungering for high-definition video
feeds directly to their computers — flow switching may make financial
sense. Chan points at the 30-inch computer monitor atop his desk in
MIT's Research Lab of Electronics. "High resolution at 120 frames per
second," he says: "That's a lot of data."
