Large Number Of Processors Research Articles

Smart traffic routing and service allocation strategy to reduce water consumption in data centers through power reduction

Due to the growth of communication networks, energy consumption in information and communication technology industries is increasing dramatically. Among these industries, data centers are operating with a large number of processors and other components, which, due to heavy processing and mass data transmission, in addition to consuming high electrical power, also cause high thermal losses. Since water cooling systems are used for cooling different parts of these centers as well as for cooling the electric power generation unit, these centers are among the major consumers of water and electricity resources. In this article, while examining the important factors affecting water consumption in data centers, useful methods are suggested to reduce the consumption of electrical energy and water. Optimizing energy consumption in data centers is possible in three parts: routing, servicing and use of cooling equipment. For all three parts, improvement methods are suggested in this article. For this purpose, an optimization problem is designed and an algorithm is presented to solve it. In the proposed solution, an energy-smart method based on SDN technology is used for routing, virtual machines equipped with the possibility of reducing power consumption in no-load conditions are used for servicing, and two types of water-cooled and air-cooled systems are used for cooling equipment. is used. The simulation results show that depending on the cooling system, the proposed method reduces water consumption between 24% and 32% compared to the case where the proposed solution is not used.

A multi-threaded approach to genotype pattern mining for detecting digenic disease genes.

To locate disease-causing DNA variants on the human gene map, the customary approach has been to carry out a genome-wide association study for one variant after another by testing for genotype frequency differences between individuals affected and unaffected with disease. So-called digenic traits are due to the combined effects of two variants, often on different chromosomes, while individual variants may have little or no effect on disease. Machine learning approaches have been developed to find variant pairs underlying digenic traits. However, many of these methods have large memory requirements so that only small datasets can be analyzed. The increasing availability of desktop computers with large numbers of processors and suitable programming to distribute the workload evenly over all processors in a machine make a new and relatively straightforward approach possible, that is, to evaluate all existing variant and genotype pairs for disease association. We present a prototype of such a method with two components, Vpairs and Gpairs, and demonstrate its advantages over existing implementations of such well-known algorithms as Apriori and FP-growth. We apply these methods to published case-control datasets on age-related macular degeneration and Parkinson disease and construct an ROC curve for a large set of genotype patterns.

Acceleration of generalized replica exchange with solute tempering simulations of large biological systems on massively parallel supercomputer.

Generalized replica exchange with solute tempering (gREST) is one of the enhanced sampling algorithms for proteins or other systems with rugged energy landscapes. Unlike the replica-exchange molecular dynamics (REMD) method, solvent temperatures are the same in all replicas, while solute temperatures are different and are exchanged frequently between replicas for exploring various solute structures. Here, we apply the gREST scheme to large biological systems containing over one million atoms using a large number of processors in a supercomputer. First, communication time on a multi-dimensional torus network is reduced by matching each replica to MPI processors optimally. This is applicable not only to gREST but also to other multi-copy algorithms. Second, energy evaluations, which are necessary for the multistate bennet acceptance ratio (MBAR) method for free energy estimations, are performed on-the-fly during the gREST simulations. Using these two advanced schemes, we observed 57.72 ns/day performance in 128-replica gREST calculations with 1.5 million atoms system using 16,384 nodes in Fugaku. These schemes implemented in the latest version of GENESIS software could open new possibilities to answer unresolved questions on large biomolecular complex systems with slow conformational dynamics.

Shape design of an artificial pump-lung using high-resolution hemodynamic simulation with high-performance computing

Accurate and fast prediction of the hemodynamics of the artificial pump-lung is critical in the design process. In this study, a comprehensive computational framework, including a sliding mesh method, a coupled free flow and porous media flow model, a hemolysis prediction method, a k−ω shear stress transport turbulence model, and solution algorithms, is introduced to accurately predict the velocity field, pressure heads, and hemolysis. The framework is used to do the shape design of an artificial pump-lung on a supercomputer. High-resolution hemodynamics simulation results are obtained and analyzed, and the parallel performance of the algorithm is studied. The numerical results indicate that the proposed framework is capable of accurately predicting the velocity field, pressure heads, and hemolysis, and the performance of the designed artificial pump-lung meets the biocompatibility requirements. Additionally, the parallel performance results demonstrate the potential of the framework to efficiently perform the design of artificial pump-lungs using a large number of processors.

Avoiding the bit-reversed ordering in parallel multi-digit multiplication based on FFT

The paper for the parallel model of computation, a modification of the method of implementing the multiplication of multi-digit integers based on the fast Fourier transform (FFT) avoiding the bit-reversed ordering is proposed. The paper researches the calculation of FFT according to the “butterfly” scheme based on decimation-in- frequency and decimation-in-time methods, an input signal with elements in direct and bit-reversed order, with an increase and decrease in the number Fourier series coefficients at each step of the "butterfly", the use of a list of Fourier series coefficients in direct and bit-reversed order. The standard FFT-based multiplication algorithm uses the same “butterfly” operation to compute the forward and inverse Fourier transforms. The paper analyzes two combinations of the FFTFDN–FFTTBN and FFTFBN–FFTTDN “butterfly” calculation schemes for calculating forward and inverse discrete Fourier transforms (DFT) in the case of implementing the multi-digit operation in parallel computational model to exclude bit-reversed permutation. A scheme for distributing calculations among four processors is proposed, in which forward and inverse Fourier transform calculations are localized within one parallel processor. The proposed modification does not reduce the computational complexity in terms of the number of complex operations, but due to the exclusion of bit-reversed permutation, the number of synchronization commands between processors and data is reduced, which reduces the algorithm execution time. The scheme can be adapted to distribute the computations among a larger number of processors. Four algorithms for implementing FFT based on decimation-in-frequency and decimation-in-time methods, an input vector with elements in direct and bit-reversed orders are presented. To check the result of the calculation, the algorithm of multiplication avoiding the steps of bit-reversed ordering was implemented in the APL programming language. An example of calculation is given in the form of a table.

Parallel multilevel domain decomposition preconditioners for monolithic solution of non-isothermal flow in reservoir simulation

In reservoir simulation, the non-isothermal flow adds a conservation of the energy equation with the involvement of a temperature variable to the complicated physics dynamics, which brings a higher nonlinearity of the corresponding PDE system and therefore exhibits significantly additional challenges on the linear preconditioner of the simulator. However, the commonly used block preconditioning techniques mostly focus on the isothermal petroleum model and do not have a specific treatment of the energy conservation equation for the non-isothermal flow. In this study, we propose and develop a family of multilevel restricted additive Schwarz (MRAS) preconditioners on parallel computers for the fully implicit solution of the non-isothermal flow in porous media. The proposed parallel reservoir simulator incorporates the restricted additive Schwarz preconditioner into a general multilevel preconditioning framework with the use of the domain decomposition technique and the multigrid method, which allows us to flexibly construct several types of multilevel Schwarz preconditioners by using the additive or multiplicative strategies. In particular, it follows the fully implicit discretization scheme for the monolithic solution, and therefore is unconditionally stable with the relaxation of the time step size by the stability condition. We numerically show that the proposed approach is highly robust and efficient for solving some non-isothermal flow problems with high nonlinearity, and good parallel scalabilities are obtained on a supercomputer with a large number of processors.

A scalable spectral Stokes solver for simulation of time-periodic flows in complex geometries

Simulation of unsteady creeping flow in complex geometries has traditionally required the use of a time-stepping procedure, which is typically costly and unscalable. To reduce the cost and allow for computations at much larger scales, we propose an alternative approach that is formulated based on the unsteady Stokes equation expressed in the time-spectral or frequency domain. This transformation results in a boundary value problem with an imaginary source term proportional to the computed mode that is discretized and solved in a complex-valued finite element solver using Bubnov-Galerkin formulation. This transformed spatio-spectral formulation presents several advantages over the traditional spatio-temporal techniques. Firstly, for cases with boundary conditions varying smoothly in time, it provides a significant saving in computational cost as it can resolve time-variation of the solution using a few modes rather than thousands of time steps. Secondly, in contrast to the traditional time integration scheme with a finite order of accuracy, this method exhibits a super convergence behavior versus the number of computed modes. Thirdly, in contrast to the stabilized finite element methods for fluid, no stabilization term is employed in our formulation, producing a solution that is consistent and more accurate. Fourthly, the proposed approach is embarrassingly parallelizable owing to the independence of the solution modes, thus enabling scalable calculations at a much larger number of processors. The comparison of the proposed technique against a standard stabilized finite element solver is performed using two- and three-dimensional canonical and complex geometries. The results show that the proposed method can produce more accurate results at 1% to 11% of the cost of the standard technique for the studied cases.

Bound-preserving inexact Newton algorithms on parallel computers for wormhole propagation in porous media

Simulating wormhole propagation during reactive dissolution of carbonates is often significantly challenging, because of the high nonlinearity of the governing equations with complex fluid physics. High resolution grids are often required to represent the complex geological heterogeneity, which demands massively parallel computers with a large number of processors. Herein, we present a parallel and scalable simulator on parallel computers for the fully implicit solution of the wormhole propagation model. Our approach is based on a family of mixed finite element methods for the spatial discretization and the implicit Backward Euler scheme for the temporal integration, to handle the combination of complicated flow physics and high resolution grids in their full complexity. Moreover, the active-set reduced-space method, as a class of bound-preserving inexact Newton algorithms, is proposed for the resultant nonlinear system of equations, to guarantee the nonlinear consistency of the fully implicit discretization in a monolithic way and to ensure the boundedness requirement of the solution. Numerical experiments are presented to demonstrate the effectiveness and efficiency of the proposed simulator for a set of heterogeneous medium problems. Large-scale results are provided to show the scalability for reservoir simulation with hundreds of millions of unknowns by using several thousand processors.

Effective Distribution of Tasks in Multiprocessor and Multi-Computers Distributed Homogeneous Systems

Nowadays, a promising is the direction associated with the use of a large number of processors to solve the resource-intensive tasks. The enormous potential of multiprocessor and multicomputer systems can be fully revealed only when we apply effective methods for organizing the distribution of tasks between processors or computers. However, the problem of efficient distribution of tasks between processors and computers in similar computing systems remains relevant. Two key factors are critical and have an impact on system performance. This is load uniformity and interprocessor or intercomputer interactions. These conflicting factors must be taken into account simultaneously in the distribution of tasks in multiprocessor computing systems. A uniform loading plays a key role in achieving high parallel efficiency, especially in systems with a large number of processors or computers. Efficiency means not only the ability to obtain the result of computations in a finite number of iterations with the necessary accuracy, but also to obtain the result in the shortest possible time. The number of tasks intended for execution on each processor or each computer should be determined so that the execution time is minimal. This study offers a technique that takes into account the workload of computers and intercomputer interactions, and allows one to minimize the execution time of tasks. The technique proposed by the authors allows the comparison of different architectures of computers and computing modules. In this case, a parameter is used that characterizes the behavior of various models with a fixed number of computers, as well as a parameter that is necessary to compare the effectiveness of each computer architecture or computing module when a different number of computers are used. The number of computers can be variable at a fixed workload. The mathematical implementation of this method is based on the problem solution of the mathematical optimization or feasibility.

Performance of parallel-in-time integration for Rayleigh B\xe9nard convection

Rayleigh–Bénard convection (RBC) is a fundamental problem of fluid dynamics, with many applications to geophysical, astrophysical, and industrial flows. Understanding RBC at parameter regimes of interest requires complex physical or numerical experiments. Numerical simulations require large amounts of computational resources; in order to more efficiently use the large numbers of processors now available in large high performance computing clusters, novel parallelisation strategies are required. To this end, we investigate the performance of the parallel-in-time algorithm Parareal when used in numerical simulations of RBC. We present the first parallel-in-time speedups for RBC simulations at finite Prandtl number. We also investigate the problem of convergence of Parareal with respect to statistical numerical quantities, such as the Nusselt number, and discuss the importance of reliable online stopping criteria in these cases.

Rendezvous algorithms for large-scale modeling and simulation

Rendezvous algorithms encode a communication pattern that is useful when processors sending data do not know who the receiving processors should be, or vice versa. The idea is to define an intermediate decomposition where datums from different sending processors can ”rendezvous” to perform a computation, in a manner that both the senders and eventual receivers of the results can identify the appropriate rendezvous processor.Originally designed for interpolating between overlaid grids with independent parallel decompositions (Plimpton et al., 2004), we have recently found rendezvous algorithms useful for a variety of operations in particle- or grid-based simulation codes when running large problems on large numbers of processors. In particular, we show they can perform well when a load-balanced intermediate decomposition is randomized and not spatial, requiring all-to-all communication to move data between processors. In this case rendezvous algorithms leverage the large bisection communication bandwidths which parallel machines provide.We describe how rendezvous algorithms work in a scientific computing context and give specific examples for molecular dynamics and Direct Simulation Monte Carlo codes which result in dramatic performance improvements versus simpler algorithms which do not scale as well. We explain how a generic rendezvous algorithm can be implemented, and also point out similarities with the MapReduce paradigm popularized by Google and Hadoop.

GOPS: efficient RBF surrogate global optimization algorithm with high dimensions and many parallel processors including application to multimodal water quality PDE model calibration

This paper describes a new parallel global surrogate-based algorithm Global Optimization in Parallel with Surrogate (GOPS) for the minimization of continuous black-box objective functions that might have multiple local minima, are expensive to compute, and have no derivative information available. The task of picking P new evaluation points for P processors in each iteration is addressed by sampling around multiple center points at which the objective function has been previously evaluated. The GOPS algorithm improves on earlier algorithms by (a) new center points are selected based on bivariate non-dominated sorting of previously evaluated points with additional constraints to ensure the objective value is below a target percentile and (b) as iterations increase, the number of centers decreases, and the number of evaluation points per center increases. These strategies and the hyperparameters controlling them significantly improve GOPS’s parallel performance on high dimensional problems in comparison to other global optimization algorithms, especially with a larger number of processors. GOPS is tested with up to 128 processors in parallel on 14 synthetic black-box optimization benchmarking test problems (in 10, 21, and 40 dimensions) and one 21-dimensional parameter estimation problem for an expensive real-world nonlinear lake water quality model with partial differential equations that takes 22 min for each objective function evaluation. GOPS numerically significantly outperforms (especially on high dimensional problems and with larger numbers of processors) the earlier algorithms SOP and PSD-MADS-VNS (and these two algorithms have outperformed other algorithms in prior publications).

Parallel multilevel restricted Schwarz preconditioners for implicit simulation of subsurface flows with Peng-Robinson equation of state

Parallel algorithms and simulators with good scalabilities are particularly important for large-scale reservoir simulations on modern supercomputers with a large number of processors. In this paper, we introduce and study a family of highly scalable multilevel restricted additive Schwarz (RAS) methods for the fully implicit solution of subsurface flows with Peng-Robinson equation of state in two and three dimensions. With the use of a second-order fully implicit scheme, the proposed simulator is unconditionally stable with the relaxation of the time step size by the stability condition. The investigation then focuses on the development of several types of multilevel overlapping additive Schwarz methods for the preconditioning of the resultant linear system arising from the inexact Newton iteration, and some fast solver technologies are presented for the assurance of the multilevel approach efficiency and scalability. We numerically show that the proposed fully implicit framework is highly efficient for solving both standard benchmarks as well as realistic problems with several hundreds of millions of unknowns and scalable to 8192 processors on the Tianhe-2 supercomputer.

Derivation and analysis of parallel-in-time neural ordinary differential equations

The introduction in 2015 of Residual Neural Networks (RNN) and ResNET allowed for outstanding improvements of the performance of learning algorithms for evolution problems containing a “large” number of layers. Continuous-depth RNN-like models called Neural Ordinary Differential Equations (NODE) were then introduced in 2019. The latter have a constant memory cost, and avoid the a priori specification of the number of hidden layers. In this paper, we derive and analyze a parallel (-in-parameter and time) version of the NODE, which potentially allows for a more efficient implementation than a standard/naive parallelization of NODEs with respect to the parameters only. We expect this approach to be relevant whenever we have access to a very large number of processors, or when we are dealing with high dimensional ODE systems. Moreover, when using implicit ODE solvers, solutions to linear systems with up to cubic complexity are then required for solving nonlinear systems using for instance Newton’s algorithm; as the proposed approach allows to reduce the overall number of time-steps thanks to an iterative increase of the accuracy order of the ODE system solvers, it then reduces the number of linear systems to solve, hence benefiting from a scaling effect.

Parallel computation for production and distribution planning

Due to the TFT LCD (Thin-Film-Transistor Liquid-Crystal-Display) manifestation into many related products, this industry has been growing rapidly along with increasing demand. To satisfy the demand, most companies have increased their production capacity and capability by increasing their number of factories in different places and causing complexity in this industry. This research develops a production and distribution planning model for the multi-stage and multi-site supply chain in the TFT LCD industry. Genetic algorithm proposed in this research to solve the problem. Maximizing capacity utilization and total profit in the supply chain are become the major performance indicator in this model. Regarding the high computational effort of genetic algorithm, then parallel computation performed. Genetic algorithm conducted in multi-processor computation using OpenMP for time efficiency. To compare the computational time, the genetic algorithm conducted in five different number of processors; 1, 2, 4, 8, and 16, to know how many processors are needed to get the optimal computational time. The result is the genetic algorithm using 4 processors has the optimal expected net profit compare to the others. The result shows that a larger number of processors doesn’t mean the computation time will become automatically faster.

System management recovery in NoC-based many-core systems

The technology nodes reduction enabled the emergence of NoC-based many-cores with dozens to hundreds of processing elements (PEs). Despite the processing power offered by a large number of processors and communication flexibility due to the adoption of NoCs, it is necessary to manage the many-core resources to ensure scalability. The execution of the management tasks requires a PE reserved exclusively to execute such actions. These processors are named managers PE–MPE. A centralized approach would induce a significant load to the MPE in large-scale systems, and a permanent fault in the MPE would compromise the entire system. The adoption of a distributed approach, organization adopted in this work, with MPEs hierarchically organized, reduces the management load, and a fault in an MPE would compromise only the PEs managed by the faulty MPE. The literature presents several fault-tolerant proposals targeting the NoC or the processors. However, there is a significant gap related to fault-tolerant methods at the system level, i.e., related to fault-tolerant techniques regarding the MPEs. The goal of this paper is to present a recovery method when an MPE became faulty, and propose a protocol to migrate the management software safely to a new PE. The method adopts task migration to release a processor if there is no processor to receive the kernel that was executing in a faulty processor. The proposal is transparent to the applications running in the many-core, with an overhead in the execution time varying between 1.5 and 1.65 ms during the management and task migration.

A Highly Parallel Multilevel Newton--Krylov--Schwarz Method with Subspace-Based Coarsening and Partition-Based Balancing for the Multigroup Neutron Transport Equation on Three-Dimensional Unstructured Meshes

The multigroup neutron transport equation is crucial for studying the motion of neutrons and their interaction with materials. Numerical simulation of the multigroup neutron transport equation is computationally challenging because the equation is defined on a high-dimensional phase space, the computational spatial domain is complex, and the materials are heterogeneous. A scalable parallel solver is required to address such a challenge. In this paper, we study a highly parallel Newton--Krylov--Schwarz (NKS) method consisting of a Newton-based eigenvalue solver, a Krylov subspace method, and a novel multilevel Schwarz preconditioner. The multilevel method is one of the most popular preconditioners for accelerating neutron transport calculations, but the construction of coarse spaces can be expensive and often unscalable when a large number of processors is used. We propose a novel matrix coarsening algorithm in which a multilevel hierarchy is constructed using a single-component matrix instead of the full matrix of the neutron transport equation. This new coarsening algorithm is referred to as “subspace-based coarsening.” Above 8,000 processors, we show a 13x enhancement in multilevel preconditioner setup time when using the subspace-based coarsening method. A partition-based balancing strategy is studied to enhance the parallel efficiency of the NKS algorithm by equalizing the work for each processor. A hierarchical mesh partitioning algorithm is employed to generate a large number of submeshes while minimizing off-node communication. We demonstrate that the proposed algorithm is scalable with more than 10,000 processors for a realistic application on three-dimensional unstructured meshes with a few billion degrees of freedom. Neutron transport calculations using the improved NKS algorithm are twice as fast as those based on the unmodified NKS solver when over 8,000 processors are employed.

Load Balancing Problem on Hyper Hexa Cell Interconnection Network

Dynamic load balancing techniques prevents computer nodes from overloading unevenly while leaving other idle. It is considered as one of the most challenging topics in parallel computing. Moreover, it is essential for increasing the efficiency of highly parallel systems especially in solving multitask problems with unpredictable load estimates. Particularly, over each processor in the parallel systems and interconnection networks. This paper focuses on developing an efficient algorithm for load balancing on Hyper Hexa Cell (HHC) interconnection network, namely, HHCLB algorithm. Basically, the Dimension Exchange Method (DEM) approach is used in this paper to construct a new load balancing approach on the network of HHC interconnections. Thus, an algorithm was introduced and simulated using java threads, where the performance of the algorithm is evaluated both analytically and experimentally. The evaluation was in terms of various performance metrics, including, execution time, load balancing accuracy, communication cost. By implementing the proposed load balancing algorithm to the HHC network, a high degree of accuracy and minimal execution time was achieved. It is important to highlight that the algorithm recorded small gap between the execution time for small number of processors and large number of processors. For instance, the algorithm achieved 0.14 seconds for balancing the load of 6 processors while 0.59 seconds for balancing the load of 3072 processors. This proves how effective the algorithm is in balancing the load for different network sizes from small to large number of processors, with a slight difference in execution time.

A New Approach to Parallel Processing

The Application Space Architecture (ASA) defines parallel processor design as an adjunct to the Instruction Set Architectures (ISA) defined by Von Neumann for single processors. Before addressing hardware, one must understand applications requiring parallel processors, and software approaches required to meet speed constraints. Parallel processor applications use a single task to run N (number-of-processors) times faster than could be done on a single processor, where Processor Utilization Efficiency (PUE) is the ratio of single processor time to parallel processor time divided by N. Such applications require processing to be split into independent elements running in parallel on a large number of processors. This requires that processors do not sit idle (current PUEs are typically 2-10%.) This sets the requirement for “independent” elements to communicate with each other without stopping. In special cases, clusters of PCs are used to perform parametric analyses running many copies of a simulation on separate computers, typically using different random numbers. Except for initial and terminal processes, these separate simulations hardly communicate while running. Thus they are easily divided into separate tasks that typically exchange information between computers passing data or messages through a top level management task. Such applications (e.g., LINPACK) are labeled “embarrassingly parallel”.

Синхронный алгоритм параллельной реконструкции трехмерных изображений на основе системы обмена сообщениями MPI

Одни из наиболее важных вычислительных задач разработка и адаптация итерационных методов для решения сверхбольших разреженных систем алгебраических уравнений. К таким вычислительным задачам приводит задача итерационной параллельной реконструкции трехмерных изображений промышленных изделий. Важно, что итерационные методы решения вычислительных задач большой размерности реализуются на параллельных структурах намного эффективнее, чем прямые методы их решения. В этой работе описан синхронный параллельный алгоритм, основанный на использовании системы MPI для решения задачи реконструкции трехмерных изображений промышленных изделий. Purpose. Currently, one of the most important tasks is the development and adaptation of iterative methods for solving ultra-large sparse systems of algebraic equations. Such computational problems are caused by the iterative parallel reconstruction of three-dimensional images of industrial products. It is important that iterative methods for solving computational problems of large size are implemented on parallel structures much more efficiently than direct methods for solving them. This paper describes a synchronous parallel algorithm based on the MPI system for solving the problem of reconstruction of three-dimensional images of industrial products. Methodology. It is important that iterative methods for solving computational problems of larger dimensionality on parallel structures are implemented more efficiently than the direct ones. Most algorithms based on direct solving methods have a significant hereditary sequential structure and require a large number of processors interactions which cannot be executed in parallel mode. Iterative methods, for the most part, require a significantly smaller number of interactions of this type and are relatively easily mapped onto parallel computational structures. Equally important, in most cases parallel implementations of classical iterative methods are more effective in terms of computational speed. The parallel execution of the algorithm is based on the distribution of the process in some way between various groups of processors. Depending on the interaction method between local processors, two different types of parallel iterative algorithms execution are distinguished: synchronous and asynchronous. In the former case, it is assumed that the processors complete the calculations and exchange all the necessary results before the start of a new iteration. The main disadvantage of synchronous parallel algorithms is that they require synchronization of iterations. This is a very difficult task, especially with large number of processors. In addition, the overall calculation speed is limited by the speed of the slowest processor. At the same time, faster processors spend most of their time in the waiting mode. But, nevertheless, the implementation of these parallel algorithms can be effectively achieved using the MPI standard. Findings. Synchronous parallel computing algorithms and program codes for threedimensional tomographic reconstruction in a conical beam were developed. Program code debugging and numerical calculations were performed on a hybrid cluster based on the OpenPOWER architecture using the MPI system. For designing a parallel threedimensional tomographic reconstruction, a voxel form of parallelism was used. Originality. Implemented parallel iterative technology of three-dimensional images reconstruction has undeniable advantages over traditional sequential iterative tomographic reconstruction. It allows reducing the time of tomographic reconstruction as many as tens of times, provides the ability to reconstruct products with sizes of 5123 to 10243 voxels with simultaneous storage of the submatrix node of the projection matrix in RAM, which eliminates the need for recalculation of matrix coefficients for each new iteration.