A team of scientists from University College London (UK) announced the development of a fundamentally new computer architecture, free from the failures and deadlocks.

It is based on algorithms borrowed from the natural chaotic systems. Computer based on an architecture (the developers call it «systemic», ie, the “system-wide”) can reprogram themselves to adapt to the actual conditions of work and to continue working in the failure of one or more components.

“Out of chaos arises order”: this statement is usually looks like a beautiful but useless in practice axiom. However, a team of scientists led by Peter Bentley (Peter Bentley) has found a way to turn an abstract formula as the basis for trouble-free computer. Take, for example such as autonomous unmanned aerial vehicles (UAV) – a flying robot with a computer can quickly change its action, coping with those obtained in the course of battle damage. There are other uses for the new computing architecture, including the development of more realistic models of the human brain.

Modern computers are ill-suited to the modeling of natural processes – the neurons or education bees. The root of these problems is enclosed in the linear nature of the operation: in general, the computer is designed for sequential execution of instructions in order. As explained by Peter Bentley: “In nature, everything is different. Natural processes are distributed, decentralized and probabilistic. They are also resistant to failures and can recover from significant damage. Computers, too, have to get this ability. ”

Modern computer as a conveyor belt: takes one instruction from memory, does it, stores the result, and then the cycle repeats. All the work is tied to one central mechanism, “the program counter.” The conveyor provide large productivity gains in time, computers have opened enormous opportunities to perform calculations and automation. At the same time, the traditional approach is ill-suited for parallel execution of tasks. Even when the user thinks that the computer is running multiple applications at once, in fact, the computer continuously switches the chain of instructions, invisible to the human eye.

Peter Bentley and his colleague, Christos Sakellariou built a new computer in which data exist on an equal footing with the instructions for processing the data. For example, the computer connects the outside air temperature to the actions in the event of heat. The results of the treatment according to the instructions are divided into pools of numerical elements, which are termed “system.”

Each received a “system” has its own memory in which the context-sensitive data. This means that a single “system” can only communicate with other such “systems”. Instead of the program counter the work of “systems” run random number generator – the generator is designed to simulate randomness of natural processes. Systems meet their team at the same time, and neither of them has any advantage over the others. The entire pool of systems of interacting simultaneously and randomly, and the calculation result is obtained as a result of all these interactions.

Verbal description of a new computer does not give reason to expect that this architecture can work. However, at a conference on the evolving system, which will be held in April in Singapore, Bentley is going to say that the system not only works, but is faster than expected. From a technical point of view, the “system” computer contains multiple copies of the instruction set, spread across multiple systems. Thus, if one system is damaged, the computer can open another clean copy to repair their shared code. Moreover, in contrast to traditional operating systems, which may fail due to lack of access to a single bit memory “system” in this case, the computer will work fine, because each of his “system” is working with its own memory.

Now the authors of “systemic” computers are hard to teach your computer to copy your own code in response to changes in the working environment. For this purpose the known machine learning algorithms.

Article based on sites New Scientist, ITProPortal and The Verge.

Advertisements