SUDDEN drastic slowdowns in Internet speed appear to follow the same rules as earthquakes. Even though we can鈥檛 yet predict earthquakes, the finding might still help Internet traffic managers detect or prevent these rumbles on the Net before they occur.
Internet traffic takes hours to build up, but a sudden slowdown can occur in just microseconds. Statistical physicist Sumiyoshi Abe at the University of Tsukuba in Japan, who was studying the phenomenon, noticed that this resembles the pattern of timescales for earthquakes. Stresses build up in rocks over years or decades, and are released in explosive bursts that last just a few seconds or minutes.
Last year, Italian scientists discovered a similarity between earthquakes and the stock market. The aftershocks that occur after a quake diminish in severity according to a distinct pattern known as Omori鈥檚 law, named after Japanese seismologist Fusakichi Omori, who discovered it in 1894. Fabrizio Lillo and Rosario Mantegna of the University of Palermo found that stock market crashes are also followed by 鈥渁ftershocks鈥� that obey Omori鈥檚 law.
Advertisement
Following that research, Abe and his colleague Norikazu Suzuki set out to test whether Internet jams resemble earthquakes too, using a computer program called 鈥淧ing鈥�. The program tests whether a network connection is active by firing a signal at a destination computer. That signal echoes back to the sender, much like sonar pulses used for echolocation under water.
The researchers measured how long it took for the pings to make their round trips, a reflection of how snarled up the Net was. They found that drastic spells of congestion were indeed followed by a series of smaller crashes, just like the tremors after earthquakes. They have submitted their findings for publication in the journal Europhysics Letters.
Larger earthquakes are also often preceded by 鈥渇oreshocks鈥�. Abe and Suzuki think this may be true for network traffic as well, and are now working on ways of checking the health of the Internet, by using smaller crashes to try to predict larger ones. But they warn that figuring out exactly when the main shock will come is likely to be extremely difficult.
Mantegna says the findings are fascinating. And although there isn鈥檛 yet a way to forecast a crash, he says Abe and Suzuki鈥檚 approach could still help prevent crashes by revealing how problems can build up to catastrophic levels.
The researchers are also keen to find out whether other complex systems, such as neural networks, might follow similar patterns.