The dissertation is devoted to research of actual problems of acceleration of calculations of big data in systems of distributed information processing by development of methods of acceleration of performance of basic operations taking into account features and properties of modern computing platforms and reduction of complexity of realization at high speed and big volume of an input data stream.
For the first time, an improved method of one-dimensional cascade of calculation of basic data processing operations is proposed, which, in contrast to the existing ones, uses constructive modules with end-to-end transfer.
It is determined that the improvement of the method provides a reduction in hardware costs by an average of 10% compared to the existing basic method.
A modification of the improved method of one-dimensional cascade implementation of the procedure for calculating basic data processing operations, which, in contrast to the previously proposed, uses structural modules of irregular type, is proposed.
It is established that the modification of the method can be used to reduce the computational complexity in the case of non-critical requirements for the regularity of the computer structure. The use of a modified method reduces the cost of computing resources by more than 50%, depending on the bit rate, compared to the existing basic method.
The method of accelerating the calculations of operations modulo for large numbers, which, in contrast to the known ones, uses group transfer chains, which allows to increase the speed of operations by 8 times compared to the implementation of the basic method.
A model of computational structures for performing operations on a variable simple module over large-digit numbers has been developed, which, unlike the known ones, allows constructing the necessary combinations of actuators to perform the same type of arithmetic instructions modulo over large-digit numbers.
An adapted algorithm for calculating the operation of multiplication and exponentiation modulo taking into account the peculiarities of its construction by the previously proposed method of one-dimensional cascade is proposed. The proposed adaptation allows to reduce the computation time by using a variable number of computationally costly operations for certain sets of input data.
The information technology of calculation of big data which, unlike existing, is based on the offered method of calculations on the module and provides acceleration of performance of computational procedures is developed.
Keywords: implementation complexity, computing means, encoding / decoding algorithms, through transfer, high-bit numbers, model of computing structures, adaptation, finite fields, productivity.
