• 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.

• 한국지능시스템학회:학술대회논문집
• /
• 한국퍼지및지능시스템학회 2003년도 ISIS 2003
• /
• pp.439-442
• /
• 2003

In order to solve the difficulties of parameter settings in SA algorithm, an improved practical SA algorithm is proposed by employing the threading techniques, appropriate software structures, and dynamic adjustments of temperature parameters. Threads provide a mechanism to realize a parallel processing under a disperse environment by controlling the flux of internal information of an application. Thread services divide a process by multiple processes leading to parallel processing of information to access common data. Therefore, efficient search is achieved by multiple search processes, different initial conditions, and automatic temperature adjustments. The proposed are methods are evaluated, for three types of Traveling Salesman Problem (TSP) (random-tour, fractal-tour, and TSPLIB test data)are used for the performance evaluation. The experimental results show that the computational time is 5％ decreased comparing to conventional SA algorithm, furthermore there is no need for manual parameter settings. These results also demonstrate that the proposed method is applicable to real-world vehicle routing problems.

• 상하수도학회지
• /
• 제22권1호
• /
• pp.49-63
• /
• 2008

The objectives of this research were to evaluate the pressurized/the main inlet water flowrate ratio which have been used as the most important parameter for operating the pump diffusion mixer until now, to suggest the alternative operating parameter and the relating criteria if the flowrate ratio was not inadequate. For the objectives of this research, computational fluid dynamics (CFD) simulation was conducted for 21 cases of flowrate ratio in full-scaled pump diffusion mixer. From the results of CFD simulation, the local velocity gradient values were calculated in each case in order to analyze the simulation results in more detail. For verifying CFD simulation, wet test was conducted. The wet test was to measure the factual coagulant dispersion distribution at a distance of 5.4m from deflector. From both results of CFD simulation and wet test, flowrate ratio was inadequate as operating parameter or criteria, on the other hand the pressurized/the main inlet velocity ratio(dimensionless) was useful in predicting the performance of pump diffusion mixer. Also, the injected coagulant could be dispersed evenly in overall cross section on the condition that pressurized/the main inlet velocity ratio(dimensionless) is over at least 20.

• Tribology and Lubricants
• /
• 제10권4호
• /
• pp.33-42
• /
• 1994

An influence of inlet pressure on the performance of sector-shaped pad thrust bearings is investigated theoretically. The optimum conditions of film thickness or the optimum positions of pivot are found through the evaluation of load capacity for all available conditions of film thickness, under the operating conditions which the thermal and pad deformation effects can be neglected. The bearing performance including the inlet pressure effects is obtained for a wide operating ranges that inertia parameter(Re$^{*}$) is up to unity, and for the various cases of pad extent angle (number of pad) and the three cases of the angle between pads. The results show that the inlet pressure has a large influence on the performance of sector-shaped pad thrust bearings. In the design of sector-shaped pad bearings, due to the inlet pressure, the optimum number of pad is varied with the operating speed and the angle between pads, and the optimum position of pivot is located toward the leading edge along with the operating speed increases.

• 한국방송∙미디어공학회:학술대회논문집
• /
• 한국방송공학회 2011년도 하계학술대회
• /
• pp.122-125
• /
• 2011

In real environment, background noise exists everywhere and degrades the performance of system. To reduce this distortion, a speech enhancement algorithm can be very useful and variety methods have been proposed. In this paper, we propose a postfilter to improve the performance of optimally modified log-spectral amplitude (OM-LSA) estimator. Proposed algorithm uses the formant postfilter to minimize perceptual distortion caused by background noise. We adjust an emphasizing parameter which is varied by spectral flatness and first reflection coefficient. The performance of the proposed algorithm is evaluated by measuring the log-spectral distance (LSD) and the perceptual evaluation of speech quality (PESQ) score. The test results show the improvement of proposed algorithm compared to conventional OM-LSA.

• 한국통신학회논문지
• /
• 제24권10B호
• /
• pp.1852-1860
• /
• 1999

본 논문에서는 음성 패킷을 전송하기 위해 새로 제시되어 표준화된 AAL2의 성능을 이산 시간 시뮬레이션 모형(discrete-time simulation model)을 통해 분석하였다. 시뮬레이션의 입력 파라메터로는 AAL2 PDU를 만들기 위한 패킷 필 지연 시간(packet fill delay), 여러 AAL2 PDU로부터 ATM 셀을 만들기 위한 셀 생성 지연(guard time), 음성 트래픽의 특성을 정의하는 버스트 정도(burstness), 그리고 사용자 수(number of channels)로 하였으며 이 값의 변화에 따른 음성 패킷의 양종단 지연 시간 변이(end-to-end delay variation)을 계산하여 각각의 파라미터가 음성 패킷의 성능에 미치는 영향을 분석하였다.

• 한국ITS학회 논문지
• /
• 제5권2호
• /
• pp.1-10
• /
• 2006

본 논문에서는 강인한 음성인식 기술의 하나인 모델 파라미터 변환 기법 중 Carnegie Mellon University(1996)에서 Moreno가 제안한 최신 VTS(Vector Taylor Series) 알고리즘을 이용하여 주어진 잡음 환경에서 실험하였다. 이러한 VTS 알고리즘의 성능평가를 위해서 기존의 잡음 처리 방법 중 CMN(Cepstral Mean Normalization) 기법을 도입하였으며, 데시벨별로 설정한 백색 잡음과 거리잡음을 환경잡음으로 주어졌을 때의 인식률을 비교하였다. 또한 기존 Moreno가 제안한 실험환경의 인식 결과와 본 논문에서의 실험결과를 비교 분석하였다. 인식 알고리즘으로는 실시간 구현이 가능한 이산HMM(Hidden Markov Model)을 사용하였다.

• Structural Engineering and Mechanics
• /
• 제58권1호
• /
• pp.121-142
• /
• 2016

The national standard being used in Turkey for suspended ceiling systems (SCS) regulates material and dimensional properties but does not contain regulations regarding installation instructions which cause substandard applications of SCSs in practice. The lack of installation instructions would potentially affect the dynamic performance of these systems. Also, the vast majority of these systems are manufactured using substandard low-quality materials, and this will inevitably increase SCS related damages during earthquakes. The experimental work presented here focuses on the issue of dynamic performance of SCSs with different types of carrier systems (lay-on and clip-in systems), different weight conditions, and material-workmanship qualities. Moreover, the effects of auxiliary fastening elements, so called seismic perimeter clips, in improving the dynamic performance of SCSs were experimentally investigated. Results show that clip-in ceiling system performs better than lay-on system regardless of material and workmanship qualities. On the other hand, the quality aspect becomes the most important parameter in affecting the dynamic performance of lay-on type systems as opposed to tile weights and usage of perimeter clips. When high quality system is used, tile weight does not change the performance of lay-on system, however in poor quality system, tile weight becomes an important factor where heavier tiles considerably decrease the performance level. Perimeter clips marginally increase the dynamic performance of lay-on ceiling system, but it has no effect on the clip-in ceiling system under the shaking levels considered.

• 한국컴퓨터정보학회논문지
• /
• 제19권11호
• /
• pp.17-24
• /
• 2014

본 논문에서는 저속으로 회전하는 유도 전동기의 베어링 결함을 검출하기 위해 음향 방출 신호와 히스토그램 모델링을 이용하는 방법을 제안한다. 제안한 방법은 정규화된 결함 신호가 구성하는 히스토그램의 포락선을 모델링하여, 부분 상관 계수와 DET(Distance Evaluation Technique) 기법을 이용하여 결함 유형별 고유한 특징을 추출 및 선택한다. 추출된 특징을 SVR(Support Vector Regression) 분류기의 입력으로 사용하여 베어링의 내륜, 외륜 및 롤러 결함을 분류한다. 최적의 분류 성능을 위해 SVR 커널함수의 매개변수를 0.01에서 1.0까지 변화시키고, 특징 개수는 2에서 150까지 변화시키면서 실험한 결과, 0.64-0.65의 매개변수와 75개의 특징 개수에서 제안한 방법은 약 91%의 분류 성능을 보였고, 또한 기존의 결함 분류 알고리즘보다 높은 분류 성능을 보였다.

• 한국건설순환자원학회논문집
• /
• 제7권1호
• /
• pp.8-15
• /
• 2019

콘크리트 구조물이 극심한 열화 환경에 노출되는 경우 열화현상이 발생하게 되는데, 대표적인 열화 현상으로는 염해에 의한 철근부식이 있다. 본 연구에서는 3가지 수준의 물-결합재 비(0.37, 0.42, 0.47) 및 2가지 수준의 플라이애시 치환율(0%, 30%)을 고려하여 재령 2년 콘크리트 시편을 대상으로 염해 저항 성능을 평가하였다. Tang's method에 준하여 촉진 염화물 확산계수를, ASTM C 1202에 준하여 통과 전하량을, KS F 2405에 준하여 압축강도를 평가하였다. 또한 기존의 제안식들과 촉진 염화물 확산계수 결과를 활용하여 각 배합의 시간의존성 확산 특성을 분석하였다. 플라이애시 혼입 배합은 OPC 배합 대비 뛰어난 내구성능을 나타내는데, 통과 전하량 평가기준에 따르면 재령 49일부터 "Moderate" 등급 이하에 포함되는 반면 OPC 배합은 재령 1년에서야 모든 물-결합재 비에서 "Moderate" 등급 이하에 포함되었다. 각 배합의 시간의존성지수를 도출한 결과 모든 배합에서 물-결합 재비가 증가함에 따라 시간의존성지수가 감소하는 경향을 나타내었다. 또한, 플라이애시 혼입 배합에서 OPC 배합 대비 1.57배~2.74배의 시간의존성지수가 평가되었다. 이는 시간의존성지수가 확산계수가 감소하는 기울기를 나타내기 때문이며, 플라이애시의 포졸란 반응에 의해 FA 배합에서 OPC 배합 대비 높은 시간의존성지수를 나타내었다.