• KIISE Transactions on Computing Practices
• /
• v.22 no.1
• /
• pp.20-25
• /
• 2016

With the advent of the era of digital big data, the Hadoop platform has become widely used in various fields. However, the Hadoop MapReduce framework suffers from problems related to the increase of the name node's main memory and map tasks for the processing of large number of small files. In addition, a method for running C++-based tasks in the MapReduce framework is required in order to conjugate GPUs supporting hardware-based data parallelism in the MapReduce framework. Therefore, in this paper, we present a face detection system that generates a sequence file for images to process big data for images in the Hadoop platform. The system also deals with tasks for GPU-based face detection in the MapReduce framework using Hadoop Pipes. We demonstrate a performance increase of around 6.8-fold as compared to a single CPU process.

• Journal of Advanced Navigation Technology
• /
• v.16 no.3
• /
• pp.495-503
• /
• 2012

As modern society rapidly changes, the field of education has also developed speedily. Since Edunet system developed in 1996, many different systems are developing continuously such as Center for Teaching and Learning, cyber home learning systems, diagnosis prescribing systems, video systems, teaching and counseling, and study management systems. However, the aforementioned systems have had not great response from the educational consumers due to a lack of interconnection. There are several reasons for it. One of the reasons is that program administrators did not carefully consider the continuity of each programs but established a brand new system whenever they need rather than predict or consider the future needs. The suitable system for smart education should be one big integrated system based on many different data analysis and processing. The system should also supply educational consumers various and useful information by adopting the idea of bigdata rather than a single sign on system connecting each independent system. The cloud computing system should be established as a system that can be managed not as simple compiled files and application programs but as various contents and DATA.

• Proceedings of the Korean Society for Agricultural Machinery Conference
• /
• 2017.04a
• /
• pp.162-162
• /
• 2017

구글에서 공개한 Tensorflow를 이용한 여러 학문 분야의 연구가 활발하다. 농업 시설환경을 대상으로 한 빅데이터의 축적이 증가함과 아울러 실효적인 정보 획득을 위한 각종 데이터 분석 및 마이닝 기법에 대한 연구 또한 활발한 상황이다. 한편, 타 분야의 성공적인 심층학습기법 응용사례에 비하여 농업 분야에서의 응용은 초기 성장 단계라 할 수 있다. 이는 농업 현장에서 취득한 정보의 난해성 및 완성도 높은 생육/환경 모델링 정보의 부재로 실효적인 전과정 처리 기술 도출에 소요되는 시간, 비용, 연구 환경이 상대적으로 부족하기 때문일 것이다. 특히, 센서 기반 데이터 취득 기술 증가에 따라 비약적으로 방대해진 수집 데이터를 시간 복잡도가 높은 심층 학습 모델링 연산에 기계적으로 단순 적용할 경우 시간 효율적인 측면에서 성공적인 결과 도출에 애로가 있을 것이다. 매우 높은 시간 복잡도를 해결하기 위하여 제시된 하드웨어 가속 기능의 경우 일부 개발환경에 국한이 되어 있다. 일례로, 구글의 Tensorflow는 오픈소스 기반 병렬 클러스터링 기술인 MPICH를 지원하는 알고리즘을 공개하지 않고 있다. 따라서, 본 연구에서는 심층학습 기법 연구에 있어서, 예상 가능한 다양한 자원을 활용하여 최대한 연산의 결과를 빨리 도출할 수 있는 하드웨어적인 접근 방법을 모색하였다. 호스트에서 수행하는 일방적인 학습 알고리즘과 달리 이기종간 심층 학습이 가능하기 위해선 우선, NFS(Network File System)를 이용하여 데이터 계층이 상호 연결이 되어야 한다. 이를 위해서 고속 네트워크를 기반으로 한 NFS의 이용이 필수적이다. 둘째로 제한된 자원의 한계를 극복하기 위한 메모 공유 라이브러리가 필요하다. 셋째로 이기종간 프로세서에 최적화된 병렬 처리용 컴파일러를 이용해야 한다. 가장 중요한 부분은 이기종간의 처리 능력에 따른 작업을 고르게 분배할 수 있는 작업 스케쥴링이 수행되어야 하며, 이는 처리하고자 하는 데이터의 형태에 따라 매우 가변적이므로 해당 데이터 도메인에 대한 엄밀한 사전 벤치마킹이 수행되어야 한다. 이러한 요구조건을 대부분 충족하는 Open-CL ver1.2(https://www.khronos.org/opencl/)를 이용하였다. 최신의 Open-CL 버전은 2.2이나 본 연구를 위하여 준비한 4가지 이기종 시스템에서 모두 공통적으로 지원하는 버전은 1.2이다. 실험적으로 선정된 4가지 이기종 시스템은 1) Windows 10 Pro, 2) Linux-Ubuntu 16.04.4 LTS-x86_64, 3) MAC OS X 10.11 4) Linux-Ubuntu 16.04.4 LTS-ARM Cortext-A15 이다. 비교 분석을 위하여 NVIDIA 사에서 제공하는 Pascal Titan X 2식을 SLI로 구성한 시스템을 준비하였다. 개별 시스템에서 별도로 컴파일 된 바이너리의 이름을 통일하고, 개별 시스템의 코어수를 동일하게 균등 배분하여 100 Hz의 데이터로 입력이 되는 온도 정보와 조도 정보를 입력으로 하고 이를 습도정보에 Linear Gradient Descent Optimizer를 이용하여 Epoch 10,000회의 학습을 수행하였다. 4종의 이기종에서 총 32개의 코어를 이용한 학습에서 17초 내외로 연산 수행을 마쳤으나, 비교 시스템에서는 11초 내외로 연산을 마치는 결과가 나왔다. 기보유 하드웨어의 적절한 활용이 가능한 심층학습 기법에 대한 연구를 지속할 것이다

• Journal of KIISE:Computer Systems and Theory
• /
• v.29 no.7
• /
• pp.401-410
• /
• 2002

Control independence has been put forward as a new significant source of instruction-level parallelism for superscalar processors. In branch prediction mechanisms, all instructions after a mispredicted branch have to be squashed and then instructions of a correct path have to be re-fetched and re-executed. This paper presents a new branch misprediction recovery mechanism to reduce the number of instructions squashed on a misprediction. Detection of control independent instructions is accomplished with the help of the static method using a profiling and the dynamic method using a control flow of program sequences. We show that the suggested branch misprediction recovery mechanism improves the performance by 2~7% on a 4-issue processor, by 4~15% on an 8-issue processor and by 8~28% on a 16-issue processor.

• Journal of the Institute of Convergence Signal Processing
• /
• v.11 no.3
• /
• pp.232-237
• /
• 2010

In this paper, a slit light laser range finding method using perspective warping calibration is proposed. This approach has an advantage to acquire relatively high accuracy, although the optical system is nonlinear. In the calibration, we detect the calibration points which are marked on the calibration panel and acquire the center position of the slit light laser in the image, which are used for computing the real positions of the slit light by using perspective warping. A calibration file is obtained by integrating the calibration data with the transition of the panel. The range data is acquired by interpolating the center position of the slit light laser to the calibration coordinates. Experimental results show that the proposed method provides the accuracy of 0.08mm error in depth range of 130mm with the low cost optical system.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2013.05a
• /
• pp.476-477
• /
• 2013

Gadget-2 is a scientific simulation code has been used for many different types of simulations like, Colliding Galaxies, Cluster Formation and the popular Millennium Simulation. The code is parallelized with Message Passing Interface (MPI) and is written in C language. There is also a Java adaptation of the original code written using MPJ Express called Java Gadget. Java Gadget writes a lot of checkpoint data which may or may not use the HDF-5 file format. Since, HDF-5 is MPI-IO compliant, we can use our MPJ-IO library to perform parallel reading and writing of the checkpoint files and improve I/O performance. Additionally, to add reliability to the code execution, we propose the usage of Hadoop Distributed File System (HDFS) for writing the intermediate (checkpoint files) and final data (output files). The current code writes and reads the input, output and checkpoint files sequentially which can easily become bottleneck for large scale simulations. In this paper, we propose Sim-Hadoop, a framework to leverage HDFS and MPJ-IO for improving the I/O performance of Java Gadget code.

• Journal of the Korean Data and Information Science Society
• /
• v.24 no.5
• /
• pp.975-987
• /
• 2013

The Hadoop system was developed by the Apache foundation based on GFS and MapReduce technologies of Google. Many modern systems for managing and processing the big data have been developing based on the Hadoop because the Hadoop was designed for scalability and distributed computing. The R software has been considered as a well-suited analytic tool in the Hadoop based systems because the R is flexible to other languages and has many libraries for complex analyses. We introduced Rhipe which is a R package supporting MapReduce programming easily under the Hadoop system, and implemented a MapReduce program using Rhipe for multiple regression especially. In addition, we compared the computing speeds of our program with the other packages (ff and bigmemory) for processing the large data. The simulation results showed that our program was more fast than ff and bigmemory as the size of data increases.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.21 no.11
• /
• pp.2067-2075
• /
• 2017

Lately big data processing is considered as an emerging issue. As a huge amount of data is generated, data processing capability is getting important. In processing big data, both Hadoop distributed file system and unstructured date processing-based NoSQL data store are getting a lot of attention. However, there still exists problems and inconvenience to use NoSQL. In case of low volume data, MapReduce of NoSQL normally consumes unnecessary processing time and requires relatively much more data retrieval time than RDBMS. In order to address the NoSQL problem, in this paper, an interworking scheme between NoSQL and the conventional RDBMS is proposed. The developed auto-mapping scheme enables to choose an appropriate database (NoSQL or RDBMS) depending on the amount of data, which results in fast search time. The experimental results for a specific data set shows that the database interworking scheme reduces data searching time by 35% at the maximum.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.53 no.1
• /
• pp.43-50
• /
• 2016

NAND flash based storage devices adopts multi-channel and multi-way architecture to improve performance using parallel operation of multiple NAND devices. However, multiple NAND devices consume higher current and peak power overlap problem influences on the system stability and data reliability. In this paper, current waveform is measured for erase, program and read operations, peak current and model is defined by profiling method, and estimated probability of peak current overlap among NAND devices. Also, system level TLM simulator is developed to analyze peak overlap phenomenon depending on various simulation scenario. In order to remove peak overlapping, token-ring based simple power management method is applied in the simulation experiments. The optimal peak overlap ratio is proposed to minimize performance degradation based on relationship between peak current overlapping and system performance.

• Journal of the Korea Institute of Military Science and Technology
• /
• v.25 no.4
• /
• pp.372-380
• /
• 2022

This study proposes a distributed parallel processing system, called the Fast Analysis System for remote sensing daTa(FAST), for large-scale satellite image processing and analysis. FAST is a system that designs jobs in vertices and sequences, and distributes and processes them simultaneously. FAST manages data based on the Hadoop Distributed File System, controls entire jobs based on Apache Spark, and performs tasks in parallel in multiple slave nodes based on a docker container design. FAST enables the high-performance processing of progressively accumulated large-volume satellite images. Because the unit task is performed based on Docker, it is possible to reuse existing source codes for designing and implementing unit tasks. Additionally, the system is robust against software/hardware faults. To prove the capability of the proposed system, we performed an experiment to generate the original satellite images as ortho-images, which is a pre-processing step for all image analyses. In the experiment, when FAST was configured with eight slave nodes, it was found that the processing of a satellite image took less than 30 sec. Through these results, we proved the suitability and practical applicability of the FAST design.