• Transactions of the Korean Society of Mechanical Engineers B
• /
• v.38 no.1
• /
• pp.71-79
• /
• 2014

A parallelization of the bi-conjugate gradient solver for the pressure equation of the CUPID (component unstructured program for interfacial dynamics) code, which was developed for analyzing the components of a pressurized water-cooled reactor, was studied in a symmetric multi-processing system. The parallel performance was investigated for three typical parallel programming models (MPI, OpenMP, Hybrid) by solving incompressible backward-facing step flow at various grid resolutions. It was confirmed that parallel performance was low when problem size was small or the memory requirement for each thread was considerably higher than the cache memory. Furthermore, it was shown that MPI was better than OpenMP regardless of the problem size, and Hybrid was the best when the number of threads was relatively small.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.3
• /
• pp.911-931
• /
• 2021

Recently, most cloud services use Docker container environment to provide their services. However, there are no researches to evaluate the performance of communication libraries for multi-GPU based distributed deep learning in a Docker container environment. In this paper, we propose an efficient communication architecture for multi-GPU based deep learning in a Docker container environment by evaluating the performances of various communication libraries. We compare the performances of the parameter server architecture and the All-reduce architecture, which are typical distributed deep learning architectures. Further, we analyze the performances of two separate multi-GPU resource allocation policies - allocating a single GPU to each Docker container and allocating multiple GPUs to each Docker container. We also experiment with the scalability of collective communication by increasing the number of GPUs from one to four. Through experiments, we compare OpenMPI and MPICH, which are representative open source MPI libraries, and NCCL, which is NVIDIA's collective communication library for the multi-GPU setting. In the parameter server architecture, we show that using CUDA-aware OpenMPI with multi-GPU per Docker container environment reduces communication latency by up to 75%. Also, we show that using NCCL in All-reduce architecture reduces communication latency by up to 93% compared to other libraries.

• Atmosphere
• /
• v.22 no.1
• /
• pp.109-115
• /
• 2012

We parallelized the CFD_NIMR model, which is a numerical meteorological model, for best performance on both of distributed and shared memory parallel computers. This hybrid parallelization uses MPI (Message Passing Interface) to apply horizontal 2-dimensional sub-domain out of the 3-dimensional computing domain for distributed memory system, as well as uses OpenMP (Open Multi-Processing) to apply vertical 1-dimensional sub-domain for utilizing advantage of shared memory structure. We validated the parallel model with the original sequential model, and the parallel CFD_NIMR model shows efficient speedup on the distributed and shared memory system.

• Journal of Communications and Networks
• /
• v.16 no.4
• /
• pp.436-446
• /
• 2014

The high intensity of research and modeling in fields of mathematics, physics, biology and chemistry requires new computing resources. For the big computational complexity of such tasks computing time is large and costly. The most efficient way to increase efficiency is to adopt parallel principles. Purpose of this paper is to present the issue of parallel computing with emphasis on the analysis of parallel systems, the impact of communication delays on their efficiency and on overall execution time. Paper focuses is on finite algorithms for solving systems of linear equations, namely the matrix manipulation (Gauss elimination method, GEM). Algorithms are designed for architectures with shared memory (open multiprocessing, openMP), distributed-memory (message passing interface, MPI) and for their combination (MPI + openMP). The properties of the algorithms were analytically determined and they were experimentally verified. The conclusions are drawn for theory and practice.

• Geophysics and Geophysical Exploration
• /
• v.21 no.4
• /
• pp.213-219
• /
• 2018

In this study, we simulated 3D wave propagation modeling using a Xeon Phi x200 processor and compared the parallel computation performance with that using a Xeon CPU. Unlike the 1st generation Xeon Phi coprocessor codenamed Knights Corner, the 2nd generation x200 Xeon Phi processor requires no additional communication between the internal memory and the main memory since it can run an operating system directly. The Xeon Phi x200 processor can run large-scale computation independently, with the large main memory and the high-bandwidth memory. For comparison of parallel computation, we performed the modeling using the MPI (Message Passing Interface) and OpenMP (Open Multi-Processing) libraries. Numerical examples using the SEG/EAGE salt model demonstrated that we can achieve 2.69 to 3.24 times faster modeling performance using the Xeon Phi with a large number of computational cores and high-bandwidth memory compared to that using the 12-core CPU.

• Journal of the Korean Operations Research and Management Science Society
• /
• v.39 no.4
• /
• pp.19-31
• /
• 2014

We study parallel processing techniques for the R programming language of high performance computing technology. In this study, we used massively parallel computing system which has 25,408 cpu cores. We conducted a performance evaluation of a distributed memory system using MPI and of a the shared memory system using OpenMP. Our findings are summarized as follows. First, For some particular algorithms, parallel processing is about 150 times faster than serial processing in R. Second, the distributed memory system gets faster as the number of nodes increases while shared memory system is limited in the improvement of performance, due to the limit of the number of cpus in a single system.

• International Journal of Naval Architecture and Ocean Engineering
• /
• v.8 no.5
• /
• pp.434-441
• /
• 2016

Numerical simulations of 1D Burgers equation and 2D sloshing problem were carried out to study numerical discrepancy between serial and parallel computations. The numerical domain was decomposed into 2 and 4 subdomains for parallel computations with message passing interface. The numerical solution of Burgers equation disclosed that fully explicit boundary conditions used on subdomains of parallel computation was responsible for the numerical discrepancy of transient solution between serial and parallel computations. Two dimensional sloshing problems in a rectangular domain were solved using OpenFOAM. After a lapse of initial transient time sloshing patterns of water were significantly different in serial and parallel computations although the same numerical conditions were given. Based on the histograms of pressure measured at two points near the wall the statistical characteristics of numerical solution was not affected by the number of subdomains as much as the transient solution was dependent on the number of subdomains.

• Geomechanics and Engineering
• /
• v.18 no.6
• /
• pp.651-659
• /
• 2019

The IB-SEM numerical method combines the spectral/hp element method and the rigid immersed boundary method. This method avoids the problems of low computational efficiency and errors that are caused by the re-division of the grid when the solids move. Based on the Fourier transformation and the 3D immersed boundary method, the 3D IB-SEM system was established. Then, using the open MPI and the Hamilton HPC service, the computational efficiency was increased substantially. The flows around a cylinder and a sphere were simulated by the system. The surface of the cylinder generates vortices with alternating shedding, and these vortices result in a periodic force acting on the surface of the cylinder. When the shedding vortices enter the flow field behind the cylinder, a recirculation zone is formed. Finally, the three-dimensional pore flow was successfully investigated.

• Proceedings of the Korea Contents Association Conference
• /
• 2014.06a
• /
• pp.19-20
• /
• 2014

• Proceedings of the Korean Information Science Society Conference
• /
• 2007.10b
• /
• pp.414-418
• /
• 2007

멀티프로세서 환경에서 OpenMP는 MPI 에 비해 병렬 프로그래밍을 쉽게 할 수 있다는 장점을 가지고 있고, OpenMP는 표준이 없는 병렬 프로그래밍 세계에서 실질적인 표준으로써 인정받고 있다. OPenMP는 대상 플랫폼에 따라 OpenMP 구현을 다르게 해야 하기 때문에 새로운 프로세서가 등장하면 그에 맞는 OpenMP구현을 만들어야 한다. 이 논문에선 다중 프로세서 시스템-온-칩 시스템인 ARM11MPCore 시스템 위에 POSIX 쓰레드에 기반하여 OpenMP 환경을 구축하고 그 성능을 측정한다.