• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제15권3호
• /
• 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.

• International Journal of Internet, Broadcasting and Communication
• /
• 제16권2호
• /
• pp.161-168
• /
• 2024

As the size of big data models grows, distributed training is emerging as an essential element for large-scale machine learning tasks. In this paper, we propose ParamHub for distributed data training. During the training process, this agent utilizes the provided data to adjust various conditions of the model's parameters, such as the model structure, learning algorithm, hyperparameters, and bias, aiming to minimize the error between the model's predictions and the actual values. Furthermore, it operates autonomously, collecting and updating data in a distributed environment, thereby reducing the burden of load balancing that occurs in a centralized system. And Through communication between agents, resource management and learning processes can be coordinated, enabling efficient management of distributed data and resources. This approach enhances the scalability and stability of distributed machine learning systems while providing flexibility to be applied in various learning environments.

• 한국습지학회지
• /
• 제22권1호
• /
• pp.15-23
• /
• 2020

본 연구에서는 다중최적화기법을 이용하여 분포형 수문모형의 매개변수 보정 과정에서 분포형의 정도가 융설과 유량의 최적화에 어떠한 영향을 미치고 있는 가를 연구하였다. 분포형 수문모형으로는 HL-RDHM를 이용하였고, 분포형 정도에 따라 집중형, 준분포형, 완전분포형 등 3개의 모형을 구성하여 최적 매개변수를 산정하였다. 유역은 108개의 격자로 구성되며, 격자별로 융설과 관련하여 15개, 유출량 관련 13개의 매개변수를 다중최적화기법인 MOSCEM를 이용하여 최적화하였다. 최적 매개변수 산정을 위하여 2004-2005년의 기상학적 자료와 융설량과 유출량 관측자료가 이용되었고, 최적화된 매개변수를 2001-2004년의 자료를 이용하여 검증하였다. 다중최적화기법 적용 결과 집중형의 경우, 초기 값에 의한 결과로 부터 RMSE 값이 융설량은 평균 35%, 유출량은 약 42% 개선되었고, 준분포형과 완전분포형의 경우는 융설량은 평균 40%, 유출량은 약 43% 정도의 RSME 값이 향상되었다. 전반적으로 집중형보다는 분포형 모형이 최적화 과정에서 융설과 유출량 예측에 더 나은 성과를 보여주었지만, 준포형과 완전분포형의 경우 최적화 성과에서 큰 차이를 보이지 않았고, 유출보다는 융설에서 분포형 정도에 따른 모형의 민감도가 더 높은 것을 확인되었다.

• 한국수자원학회:학술대회논문집
• /
• 한국수자원학회 2005년도 학술발표회(1)
• /
• pp.793-794
• /
• 2005

• 한국음향학회지
• /
• 제37권6호
• /
• pp.437-442
• /
• 2018

본 논문에서는 환형배열을 이용하여 단일음원이 코히어런트 분산 분포를 가지는 경우 닫힌 형태로 음원의 위치를 추정하는 기법을 제안한다. 음원이 다중경로를 거쳐 센서에 도달하는 경우 분산음원으로 보이며 이때 음원의 위치는 대표 방위, 대표 고각, 대표 방위의 분포, 대표 고각의 분포 네 가지 변수로 표현될 수 있다. 이러한 경우 DSPE(Distributed Source Parameter Estimator) 기법과 같은 탐색 기법으로 네 변수를 찾기 위해서는 매우 많은 탐색과정을 필요로 한다. 본 논문에서는 빠른 위치 추정을 위해 센서간의 상관함수와 최소자승기법을 이용하여 닫힌 형식으로 대표 방위와 고각을 추정하는 기법을 제안한다. 특히 음원이 대표적인 분포 모델인 가우시안 분포를 따를 경우 방위와 고각의 표준편차 또한 닫힌 형식으로 추정한다. 시뮬레이션에서는 DSPE 기법과 비교하여 제안 기법의 타당성을 확인하였다.

• 전기학회논문지
• /
• 제60권2호
• /
• pp.279-284
• /
• 2011

Harmonic distortions in current waveform may cause significant problems in electric power system facility and operation. This paper presents an adaptive parameter estimation method to detect harmonic current components caused by nonlinear loads. In addition, a coordination strategy for multiple inverter-interfaced distributed generators to compensate the harmonic currents is discussed. The coordination strategy is realized by distributing the harmonic compensation participation index to individual distributed generators. The harmonic compensation participation index can be determined by the amount of remaining power generation capacity of each distributed generator. Simulation results based on switching-level inverter models show that the proposed harmonic detection method has good performance and the coordination strategy can improve harmonic problems efficiently.

• Journal of applied mathematics & informatics
• /
• 제8권1호
• /
• pp.199-212
• /
• 2001

We consider a k-out-of-n system with repair under T-policy. Life time of each component is exponentially distributed with parameter $\lambda$. Server is called to the system after the elapse of T time units since his departure after completion of repair of all failed units in the previous cycle or until accumulation of n-k failed units, whichever occurs first. Service time is assumed to be exponential with rate ${\mu}$. T is also exponentially distributed with parameter ${\alpha}$. System state probabilities in finite time and long run are derived for (i) cold (ii) warm (iii) hot systems. Several characteristics of these systems are obtained. A control problem is also investigated and numerical illustrations are provided. It is proved that the expected profit to the system is concave in ${\alpha}$ and hence global maximum exists.

• 한국자동차공학회논문집
• /
• 제15권6호
• /
• pp.177-182
• /
• 2007

In this paper, automotive hydraulic pipeline systems are modeled in which a straight blocked pipe, two pipes with sudden expansion or contraction are connected in series and terminated with a chamber. The frequency response characteristics of these composite pipeline systems are investigated experimentally. The theoretical analysis for various pipe configurations is base on transfer matrix method with frequency dependent viscous friction distributed parameter pipeline model. The gain and phase of transfer functions are included for comparison with experimental results. There is close agreement between the results of experimental and theoretical determination of pressure response in automotive hydraulic pipeline systems.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 1994년도 Proceedings of the Korea Automatic Control Conference, 9th (KACC) ; Taejeon, Korea; 17-20 Oct. 1994
• /
• pp.319-323
• /
• 1994

In this paper, we discuss a design method of iterative learning control systems for parabolic linear distributed parameter systems(DPSs). First, we discuss some aspects of boundary control of the DPS, and then propose to employ the Karhunen-Loeve procedure to reduce the infinite dimensional problem to a low-order finite dimensional problem. An iterative learning control(ILC) for non-square transfer function matrix is introduced finally for the reduced order system.

• 대한전기학회논문지
• /
• 제35권3호
• /
• pp.95-101
• /
• 1986

This paper describes two methods for analyzing distributed parameter systems (DPS) via Walsh series expansions. Firstly, a Walsh-Galerkin expansion approach technique (WGA) introduced by S.G. Tzafestas. is considered. The method which is based on Galerkin scheme, is well established by using Walsh series. But then, there are some difficulty in finding the proper basic functions at each systems. Secondly, a double Walsh series approach technique (DWA) is developed. The essential feature of DWA propoesed here is that it reduces the analysis problem of DPS to that of solving a set of linear algebraic equation which is extended in double Walsh series.