VIRUSES are common on the Internet, but now it鈥檚 the turn of bacteria, says
British Telecom. But rather than being a menace, the organisms BT has in mind
have been designed to help make the Net faster. The idea is to mimic bacterial
sex to 鈥渆volve鈥� a more efficient network.
Ian Marshall at BT鈥檚 research centre near Martlesham, Suffolk, believes that
the volume of Internet traffic is growing at an unsustainable rate. 鈥淚t鈥檚
important that the network is able to adapt continuously,鈥� he says. BT鈥檚
solution is to let it manage itself.
Rather than having software oversee the entire network, Marshall and his
colleague Chris Roadknight designed the junctions of a network so that each
could be run by a piece of autonomous software. Like bacteria, they had no
central knowledge or overall control, yet Marshall says they were able to run
the network very efficiently.
Advertisement
Information sent over the Internet is broken up into discrete packets which
are stored in a 鈥渂uffer鈥� memory before being forwarded to the next leg of their
journey. Ideally, buffers are long. But the more information the packets hold,
the slower they move. 鈥淚f a packet arrives at the buffer and it is full, then it
tends to get thrown away,鈥� says Marshall. This means some information will be
lost.
What is needed is the ability for buffers to know what kind of information
they are passing and adapt their length according to the need. To do this using
traditional techniques will become impossible as the Internet grows. So BT
turned to nature in the form of 鈥済enetic algorithms鈥�. These are biologically
inspired programs commonly used to help design鈥攐r evolve鈥攖hings that
people find difficult.
GAs mimic natural selection by treating strings of data like genetic
material. These strings can be combined and mutated to produce offspring whose
fitness is then evaluated. The best are 鈥渂red鈥� to produce more offspring. This
is repeated until the best design has evolved. The problem BT found was that GAs
are too slow for real-time applications, since they require the evaluation to
take place over many generations. So Marshall and his colleagues proposed a
different biologically inspired solution.
Bacteria don鈥檛 have sex in the traditional sense, they reproduce using a
technique known as plasmid migration. 鈥淭hey wander around and bump into each
other and say `hi, do you want some of my genetic material鈥�,鈥� explains Inman
Harvey, an artificial intelligence researcher at the University of Sussex. By
exchanging plasmids (rings of genetic information) through their cell membranes
with neighbouring cells, successful genes can pass quickly through a colony and
allow it to adapt rapidly to its environment. Random mutations produce new
genes, with successful genes flourishing and ineffective ones quickly dying off.
In a network these successful genes will enable the buffers to be optimised to
carry packets of information as efficiently as possible. According to Harvey,
the technique could open up many opportunities.
鈥淲ith evolution, traditionally you only reap the benefits after many
generations, whereas here you have the three aspects of
evolution鈥攈eredity, variety and selection鈥攂ut without anyone having
to die,鈥� he adds.