The Future Of Computers Essay Research Paper — страница 2

  • Просмотров 242
  • Скачиваний 5
  • Размер файла 16

drawback of share memory is the memory itself. Fighting congestion and bottlenecking (too many processors trying to access and manipulate data at the same time) is difficult and expensive. That’s why in some systems all the processors are allocated their own memory in a scheme called single-instruction/multiple data processing. In these systems data is sent at the beginning of the computations to all the processors involved. During computation however many processors will need to share data with others or receive additional input. Therefore, for efficient rapid memory accessing and data distribution, the quickest shortest connections for communication must be attained, amount of interpathways necessary must be minimized and the right scheme for distributed memory to the

application at hand must be matched. N-Dimensional Cube Architecture.The n-dimensional cube also known as the hypercube is a multidimensional shape of n dimensions, hence is other name the n-cube. To create a one dimensional plane we connect two points in space, for a two dimensional plane we take two lines and connect each of the two points to the corresponding points on the other line to form a square. For three dimensions we take the two squares and connect their corresponding points to form a cube. For the 4th dimension we do the something and so on and so forth. So in general and n-dimensional cube will have 2n nodes because there would be a processor at each connection. So in doubling the number of connections (i.e. each additional dimension doubles the number or

connections) we get a more efficient way of connecting many processors together. In a four dimensional hypercube the longest path connecting two processors four communication lines. The unique structure precludes the need to wire every single processor to every other and still ensure an efficient data path. So in and n-dimensional hypercube the longest path between any two nodes is n. Current practical limitations place hypercubes at a maximum of 16 dimensions were 65,536 processors are interconnected. Their theoretical peak speeds are over 262 billion floating-point operations per second. Dataflow Architecture.Serial computers are limited to following a list of instructions step-by-step. Parallel computers free researchers to experiment, innovate and imagine radically different

approaches to programming control. One of these computing strategies is the concept of dataflow, pioneered by Jack Dennis at MIT. Here data is sent to and from a processor as is needed or as solutions become available. A properly running dataflow machine should be able to maintain a constant stream of data flowing toward the solution.Dataflow machines are roughly like a railroad switchyard. All the processors, about 100 due to coarse-grained nature of most dataflow machines, are connected to this switchyard. Before computation, an instruction for each processor results. This new data is tagged with information about where it should go and how it should be used. The central yard switch will able to read these “tags” and determine their appropriate routes. Nodes in a dataflow

machine operate until all there necessary data arrives. Unused nodes remain idle. But by waiting for required data, the dataflow machine eliminates the danger of clashing with ongoing work of other processors and creating a bottleneck. When the waited upon data arrives it is tagged and sent back through the network again. So the user only needs to specify instructions to be completed and allocate the right information to the processors once, and then let the flow of data reach a solution itself without having to be burdened with exact steps, procedures, r details of execution. As the dataflow grows, it has difficulty in dealing with its increased size and complexity. There are more communications, wiring, processors, and expenses. The cost becomes impractical which is why most

dataflow machines are coarse-grained. If dataflow architecture is to become the architecture of choice for future computers it must be able to cope with the large database of real-world applications and artificial intelligence. Optical ComputersLight speed (the fastest anything currently known to us can go) is a constant one hundred eighty six thousand miles per second. Researchers the world over hope to harness this speed. Optical computers promise speeds much faster than today’s computers can achieve. While silicon transistors have a projected speed limit of an operation ever 50 pico seconds, or a trillionth of a second, optical transistors might reach speeds of up to an operation ever femtosecond, or quadrillionth of a second (a million-billionth of a second). If successful