快猫短视频

IT鈥橲 ALL there at your fingertips鈥攙ideo, music, text, even smutty
pictures. And, boy, do you lap it up. You download anything and everything from
the Web. Most of it may be frivolous, but it鈥檚 still all new and fun. You don鈥檛
really need to hear a presidential greeting when you visit the White House Web
site, but you download it anyway. And why not? It costs nothing鈥攐r next to
nothing鈥攖o pluck an extra page from the Web.

Well, sorry to spoil the fun, but there is a cost. The result of this endless
search for good surf is that the Internet is grinding to a halt. The amount of
electronic traffic has become so monstrous that nobody can rely on the Internet
for urgent e-mails. Without realising it, all Web users are modern players in an
old social dilemma known as the tragedy of the commons. By blindly acting in
their own interests they are spoiling a valuable common resource. The dilemma
dates back to Medieval Britain, where thoughtless villagers bought too many
animals and overgrazed their common land, but it applies just as well to the
Net.

The intervening centuries have seen few new ideas for solving the problem.
But ironically enough, the Internet turns out to be an ideal place to understand
and control the very self-interested behaviour that is bringing it to a halt.
This endeavour is spawning a whole new field of research that brings together
physicists, economists and mathematicians. The payoff could be not only a
much-needed easing of traffic on the Net but also a better understanding of the
global impact of group dynamics.

When physicists want to calculate the impact of groups of interacting atoms
or molecules, they employ statistical mechanics, which was developed late last
century by the Austrian physicist Ludwig Boltzmann and others. They found that
while the motions and speeds of individual particles in a gas were
unpredictable, their collective motion determined the pressure, temperature and
other properties of the gas. To calculate these macroscopic properties from the
particles鈥� microscopic activity, Boltzmann and his contemporaries created a raft
of new statistical techniques.

But it鈥檚 not clear how well statistical mechanics translates to humans.
鈥淓xpectation of the future is why people are different from rocks and bottles,鈥�
says Bernardo Huberman, a physicist at Xerox Palo Alto Research Center (PARC) in
California and one of the leaders in fusing physics and economic theories. 鈥淭his
determines the `forces鈥� which act upon us. We commit to action based upon our
projection about what we think is going to happen.鈥�

Mathematics has a different tool for dealing with people鈥檚 actions: game
theory. One game theory puzzle, called prisoner鈥檚 dilemma, is a scaled-down
version of the tragedy of the commons. In this puzzle, two criminals have to
decide whether to give evidence against the other or keep silent. If they both
keep quiet they both get off, but if one implicates the other, they both go to
jail, but the 鈥渄efector鈥� gets a reduced sentence. They must decide whether to
collaborate and keep silent or act alone. 鈥淭hey should hold out, but they
don鈥檛,鈥� says Steven Durlauf, a professor of economics at the University of
Wisconsin. Both end up worse off than if they had cooperated.

In an effort to model traffic on the Internet, Huberman has married game
theory with the equations of statistical mechanics. Just as statistical
mechanics depends on the assumption that gas particles have random speeds and
directions, Huberman supposed that people decide whether to surf the Net or log
off depending on the rules of game theory. Cooperators log off, while defectors
stay online. Huberman proposed that his equations should give the collective
impact on the Net of many people鈥檚 actions, just as the equations of statistical
mechanics reveals the macroscopic properties of a gas from the activity of
millions of particles.

Solving the equations gives a pattern of Net traffic known as a 鈥渓ognormal鈥�
distribution: a bell curve which has been squashed to one side
(see Diagrams).
What this shows is that most of the time only a few people are demanding
data. But occasionally lots of people try to download data at the same time and
jam the works up. These spikes of congestion, which appear at random and last a
few seconds or so, are Net 鈥渟torms鈥�.FIG-21225102.jpg

Computer simulations, which modelled the decisions of hundreds of users,
backed up Huberman鈥檚 analysis. Traffic was usually low, but storms swirled
through the system at random. Then, as people logged off in frustration, the
jams died down quickly. 鈥淭he bursty behaviour is fascinatingly complicated,鈥�
says Huberman. And as the Net grows, he warns, the tragedy of the commons will
drag the global network to a halt.

The next step was to test the theory, and one great advantage of the Net is
that collecting data from it is simple. Huberman and his colleagues sent tiny
packets of data, called ping packets, back and forth between PARC鈥檚 computers
and others in the US and Britain. Sure enough, though most packets got through
in a reasonable time, there were short storms of congestion. The pattern of
traffic fitted the lognormal distribution. So Huberman can now calculate the
pattern of congestion on the Net using the behaviour of people faced with a
social dilemma as his starting point. He hopes this insight will help to defeat
the increasing congestion on the Net.

Hit them where it hurts

Calculating the collective impact of people鈥檚 behaviour is one thing,
changing it is quite another. Economists say that the best route is to hit them
where it hurts鈥攊n the bank balance. This means changing how people pay for
services on the Net. But how should they be changed so as not to ruin the
egalitarianism of the Net?

To decide on the best scheme, economists must understand the forces that
shape the market. But experiment is lacking. 鈥淓verything has been theoretical to
date,鈥� says economist Hal Varian, the dean of the School of Information
Management and Systems at the University of California at Berkeley.

But for Varian, like Huberman, the Net seems to offer great experimental
possibilities. 鈥淲hat we鈥檙e trying to do is measure the demand for different
dimensions of quality of service,鈥� he says. At Berkeley, 150 people have been
given access to the Net鈥攚ith a catch. They have to choose how good the
service is. They can select different bandwidths, how long they want to wait to
get connected to the Net, or how often their system randomly garbles data. The
better the service, the more it costs. Once the study is finished, economists
should be able to see what sort of pricing schemes will change people鈥檚
behaviour, and so ease Net traffic.

A number of pricing schemes are already on the cards. In one such scheme,
鈥淧aris metro pricing鈥�, the Net is divided into two separate systems, labelled
class one and class two. The systems are identical except that class one costs
more so fewer people use class one, thereby guaranteeing a faster service. By
contrast, 鈥減roportionally fair pricing鈥� uses a single Internet, but there is a
sliding scale of prices: the more you pay, the better the service.

But these schemes are not ideal. They take no account of the fact that some
types of data contribute more to the Net鈥檚 traffic problems than others. These
pricing schemes are 鈥渏ust stopping disaster鈥�, says Frank Kelly, a mathematician
from the University of Cambridge. 鈥淚f the Net is to survive,鈥� he argues, 鈥測ou
need ways of refining the idea of making use of resources.鈥�

A more sophisticated approach to easing traffic on the Net is emerging, once
again, from physics. In 1948, Claude Shannon, a mathematician working at Bell
Labs in New Jersey, forged an odd link between computer communications and
thermodynamics. He defined the entropy of a stream of digital bits in a similar
way to that used by physicists to describe the entropy of a clump of matter.
Entropy is a measure of disorder. One can think of it as the difficulty in
predicting where an atom will be at some time in the future and how it will be
moving. A crystal, in which the atoms are all constrained by their neighbours,
has lower entropy than a gas, in which the atoms are free to move. Similarly, if
it is easy to predict where the next 1s and 0s will appear in a stream of bits,
then the stream has low entropy. A random stream has high entropy.

This link between the physical world and information theory has weird
consequences. For example, commonplace terms, such as temperature, can be
translated so they apply to data. Heat up a material and its atoms move about
faster, so its entropy increases. Likewise, the more 鈥渂ursty鈥� a stream of data
is, the harder it is to predict the next burst, so its entropy is higher. This
is a 鈥渉ot鈥� stream of data. Researchers from the Dublin Institute of Advanced
Studies, the University of Cambridge and Telia, a Swedish telecommunications
company, are hoping that a data stream鈥檚 temperature will give an idea of how
difficult it is to transmit.

鈥淭emperature measures the impact of data on the system,鈥� says Simon Crosby,
one of the Cambridge computer scientists working on the project. A steady stream
of data is easy for even a simple network to cope with. But when a data storm
hits, the network must be able to cope with wildly fluctuating rates and even
overloads. By charging more to send high-temperature data, Net providers should
reduce the amount of bursty traffic, relieving the stress on their Internet
connections. 鈥淓ntropy models are quite sophisticated models that may be moved to
over several years,鈥� says Kelly.

Today, people surf the Net by reading a Web page for a while before
downloading another. This is inherently bursty. So charging people on an
entropy-based scale could force people to work differently online. They would
need new browsers, capable of smoothing out surges in data flow.

So science applied to cyberspace may help to protect the commons of the Net
from ruin. But this strange fusion may also yield other benefits. Huberman is
already using the Net as a laboratory to explore everyday social interactions.
鈥淚鈥檓 very interested in how informal networks of people form and coalesce,鈥� he
says. By analysing the way groups鈥攕uch as a set of schoolchildren in
Arizona鈥攕ent messages to one another, Huberman built a statistical model
of how people interact at work and at parties.

鈥淵ou can discern patterns of behaviour. If you walk into a party and there
are only five people, you鈥檒l talk to all of them,鈥� says Huberman. 鈥淏ut in a
large party, you鈥檒l find a few people to talk to who are the most interesting to
you.鈥� Studying group behaviour may never be the same again.

* * *

Jam buster

JUST as canny traders make a lot of money on the volatility of stock markets,
Bernardo Huberman, a physicist at of Xerox PARC has designed a superfast Web
browser that exploits Internet storms.

To minimise the possibility of financial disaster, traders 鈥渄iversify鈥�,
choosing a variety of stocks from different market sectors. If one crashes, it
will more than likely be balanced by a good performer in the portfolio. Huberman
applied this approach to speed up the way a computer coloured in a graph. The
problem was to colour a collection of connected dots so that no two adjacent
dots had the same colour. In general, this takes quite a while to solve.

Huberman set up the computer to run several copies of a graph-colouring
program at the same time, each one beginning at a different point on the network
of dots. One copy received most of the processor鈥檚 attention, while the others
ticked away slowly in the background. Once in a while, a background copy struck
lucky and found an answer quickly鈥攂ecause it started from a better point.
He calculated that these occasional jackpots more than compensated for the short
time the computer spent away from the main copy of the program.

An extra refinement is that if the main copy looks like getting now here, the
computer can cut its losses and reinvest its processor time in a background
copy. With this combined approach, Huberman solved the problem in an average of
one-tenth the time a single program took, yet using only the same computing
power.

Huberman then applied this method to the Net. A Web browser, for instance,
could activate several copies of a program to download a large graphic. Since
the bits that make up a message on the Net often take different routes, one
program might strike it lucky and find an open route. Huberman鈥檚 understanding
of the statistical nature of Net traffic allows him to optimise this process.
The bursty nature of the traffic ensures that while many programs will encounter
spikes of congestion, at least some will find a time slot free of heavy traffic.
鈥淵ou can `ride鈥� the instabilities and compress the time it takes to send a
message,鈥� says Huberman.

Of course, the paradox here is that if everyone used Huberman鈥檚 fast browser,
the Net would soon grind to a complete halt. All the lulls in traffic would
quickly be filled by background programs looking for an open road. Back to
square one.