Earticle

A 4Gb/s transceiver with the CTLE is presented. The proposed CTLE can recover the attenuated data up to 7dB by changing the digital codes which are received from the digital control logic. The proposed transceiver was designed and fabricated in 180-nm CMOS technology and consumes 60.37mW.

최신 GPU의 구조는 주요한 각 그래픽 연산별로 특화된 유닛을 사용하는 스트림 아키텍처를 넘어서서 여러 개의 동일한 Programmable한 유닛들의 집적화를 통해 범용 연산의 실행이 용이한 형태로 발전하고 있다. 또한 이전의 GPU에서 연산을 수행하기 위해 그래픽에 특화된 API를 사용하던 환경 대신, 최신의 GPU에서는 프로그래머가 보다 직접적으로 GPU를 제어할 수 있고 사용이 편한 소프트웨어 환경들이 개발되고 있다. 이러한 하드웨어와 소프트웨어의 발전은 GPU를 활용한 범용 컴퓨팅(GPGPU)을 보편화 시키고 있다. 이러한 추세에 맞추어 본 논문에서는 최신의 GPU와 소프트웨어 개발 환경을 활용하여 계산 요구량이 높은 금융파생 상품 모델링 응용 프로그램을 병렬화하고, 성능을 최적화하는 연구를 수행한다. 또한 GPU의 성능 측정 결과를 CPU만 사용하는 경우와 비교하여 분석한 결과를 제시한다. 실험 결과 GPU를 활용한 경우 190배 이상의 큰 성능 향상을 얻을 수 있었다.

The architecture of the latest GPU has surpassed the previous application-specific stream architecture. Thishas led to an architecture consisting of a number of uniform programmable units integrated on the same chipwhich facilitate the general-purpose computing beyond the graphic processing. With the multipleprogrammable units executing in parallel, the latest GPU shows superior performance. Furthermore,programmers can have a direct control on the GPU pipeline using easy-to-use parallel programmingenvironments, whereas they had to rely on specific graphics API’s in the past. These advances in hardwareand software make General-Purpose GPU (GPGPU) computing widespread. In this paper, using the latestGPU and its software environment, we parallelize a computationally demanding financial application andoptimize its performance. We also analyze the performance results compared with those obtained using CPUonly. Experimental results show that GPU can achieve a superior performance, grater than 190x, comparedwith the CPU-only case.

계산과학분야에서 고성능컴퓨팅(HPC)을 활용하기 위해서는 프로그래밍, 알고리즘, 자료구조 등 컴퓨터과학의 지 식들과 기술들을 배워야 한다. 본 논문에서는 계산과학분야의 IT교육현황 조사와 설문조사를 통해 고성능컴퓨팅 교육을 개선시키기 위한 정책 방향을 제안하는데 있다. 이를 위해 국내 대학의 물리학, 화학, 생명과학, 지구과학분야의 전공과목 중에서 IT관련 과목 현황과 사용자들의 국내 고성능컴퓨팅 교육에 대한 인식을 조사하였다. 그 결과 계산과학분야의 IT과 목비율은 응용 전공과목에 비해 매우 낮았다. 대학의 교육 요구도는 높게 나왔지만, 대학의 교육 제공 수준은 제일 낮게 나왔다. 또한 대부분의 사용자들은 독학으로 필요한 지식과 기술들을 습득한 것으로 조사되었다. 즉 대학의 역할이 가장 시급하고 중요하며 전문기관과 온라인교육의 역할도 중요하다고 확인하였다.

In order to utilize HPC in Computational science, It is necessary to learn the knowledge and skills of computer science such as programming, algorithms and data structure. In this paper, we investigate IT education status in Computational science and propose policy directions to improve the HPC education through user survey. To do this, we surveyed the current state of IT subjects among major subjects in physics, chemistry, life sciences, and earth science in domestic universities and surveyed the users' Recognition of HPC education. As a result, the ratio of IT subjects in Computational science was very lower than the ratio of major domain subjects. Despite the high educational needs of universities, the educational level of universities was the lowest. Most users have learned the necessary knowledge and skills through self-study. We recognized the role of the university is the most urgent and important, and the role of professional institutions and online education is also important.

반도체 집적 기술의 발전으로 중앙처리장치 및 저장장치가 소형화되고 성능이 빠르게 발전되면서 고성능 컴퓨팅(High Performance Computing) 분야에서 인터커넥션 네트워크가 전체 시스템의 성능을 결정하는데 더욱 중요한 요소가 되고 있다. 본 논문에서는 고성능컴퓨팅 분야에서 사용되는 인터커넥션 네트워크 기술 동향을 분석하였다. 2017년 6월 기준 슈퍼컴퓨터 Top 500에서 가장 많이 사용하고 있는 인터커텍트는 인피니밴드이다. 최근 이더넷은 40/100Gbps 기가비트 이더넷 기술의 등장으로 인피니밴드 다음으로 높은 점유율을 보이고 있다. 지연(latency) 성능이 인피니밴드에 비해 떨어지는 기가비트 이더넷은 비용 대비 효율을 중시하는 중형급 데이터 센터에서 선호하고 있다. 또한 고성능을 요구하는 최상위 HPC 시스템들은 기존의 이더넷, 인피니밴드 기술에서 벗어나, 자체적인 인터커넥트 네트워크를 도입하여 시스템의 성능을 극대화 하는 노력을 하고 있다. 향후 고성능 인터커넥트 분야는 전기 신호기반 데이터 통신에서 한 단계 도약하여, 빛으로 데이터를 주고받는 실리콘 반도체 기반 광송수신 기술이 활용될 것으로 예상된다.

With the development of semiconductor integration technology, central processing units and storage devices have been miniaturized and performance has been rapidly developed, interconnection network technology is becoming a more important factor in terms of the performance of high performance computing system. In this paper, we analyze the trend of interconnection network technology used in high performance computing. Interconnect technology, which is the most widely used in the Supercomputer Top 500(2017. 06.), is an Infiniband. Recently, Ethernet is the second highest share after InfiniBand due to the emergence of 40/100Gbps Gigabit Ethernet technology. Gigabit Ethernet, where latency performance is lower than InfiniBand, is preferred in cost-effective medium-sized data centers. In addition, top-end HPC systems that demand high performance are devoting themselves from Ethernet and InfiniBand technologies and are attempting to maximize system performance by introducing their own interconnect networks. In the future, high-performance interconnects are expected to utilize silicon-based optical communication technology to exchange data with light.

GPU는 저비용 고효율의 프로세서로 각광을 받고 있으며, 이질적 고성능 컴퓨팅 시스템의 주계산 자원으로 채택되고 있다. 본 논문에서는 GPU 기반의 고성능 클라우드 컴퓨팅을 위해 개발된 확장형 OpenStack을 서술한다. 확장형 OpenStack은 OpenStack에서 GPU 사용이 가능한 가상 머신의 생성 및 관리가 가능하도록, 가상 머신 간의GPU 공유 스케줄러와 GPU 인지형 Nova 스케줄러를 제공한다. 확장형 OpenStack과 Rodinia 벤치마크를 이용한 실험에서 GPU 가상화로 인한 오버헤드가 2%내에 불과함을 보여주고 있다. 이것은 OpenStack이 이질적 고성능 클라우드 컴퓨팅에도 성공적으로 적용 가능함을 의미한다.

GPU's are getting the spotlight in the chip processor market due to low power consumption and high efficiency, and increasingly adopted as a main computing resource in heterogeneous high performance computing systems. This paper describes OpenStack extension for heterogeneous high performance computing on the cloud. The extended features are a coarse-grained GPU scheduler and a GPU-aware Nova scheduler developed for creating and managing GPU-enabled virtual machines with OpenStack. The experiments by using Rodinia benchmark on OpenStack extension shows that the overhead due to GPU virtualization is within 2%. This means that OpenStack is successfully applicable to the heterogeneous high-performance cloud computing.

The high speed CPU based routers currently in use could not handle the massive data required for real-time multimedia communication. Graphics processing units (GPUs) offer an appreciable alternative due to high computation power which results from their parallel execution units. This paper presents the implementation of the Dijkstra’s link state IP routing algorithm using GPU. Experimental results show that the proposed GPU-based approach outperforms the same sequential CPU-based implementation in terms of execution time for the same dense graph. In addition, the proposed GPU-based approach reduces about 99% energy consumption over the CPU-based implementation.

High Performance Computer (also called "supercomputer") is a computer which can process the complex and large-scale operation in computational science. In general, High Performance Computer represents the top 500 computers which is based on the computer performance in the world. After the successful commercialization of CDC6600 which has the 9MFLOPS performance in 1964, TFLOPS and PFLOPS High Performance Computer have been developed. High Performance Computer is the public resource which can be developed and provided by government. For the efficient application of High Performance Computer, government has to support the related activities such as research, resource allocation, and professional manpower training. To build a basis of the development of High Performance Computing can improve the quality of life and develop the national economy. For this reason, Korean government preceded the establishment of law for national supercomputing promotion from 2009. As a result of this effort, Korean government enacted the 'National Supercomputing Promotion Act' in 2011. This study provides the domestic and international trend of High Performance Computer, and analyses the major contents of National Supercomputing Promotion Act. This effort will be helpful to propose the strategy in order to provide more efficient service as a momentum of national supercomputing center.

Traditionally computational scientists have used supercomputers to solve their scientific problems using multi-processors and large-scale shared memory, and scientific communities have also used Grid computing for scientific projects using large-scale computation and data resources. Recently Cloud computing is a emerging infrastructure to be considered for scientific applications and several institutes applied this to their projects. But there is no sure method for computational scientists to select a environment to fit their researches though diverse researches using these infrastructures are conducted in a variety of science domains. In this paper we describe three infrastructures for scientific researches through literature reviews, and we deliberate on what factors are considered to select one infrastructure to solve scientific problems from a scientists’ point of view. And then, analytic hierarchy process is introduced as an approach for choosing a proper infrastructure, and we propose a model to choose a appropriate infrastructure reflected by various aspects of computational science’s properties via this approach. Thus this paper offers a novel viewpoint to scientists when they choose a high performance computing environment for their researches.

In distributed high-performance computing environments, resource reservation mechanism is an effective approach to provide desirable quality of service for large-scale applications. However, conventional reservation service might result in lower resource utilization and higher rejection rate if it is excessively applied. Furthermore, redundant reservation policy has been widely applied in many practical systems with aiming to improve the reliability of application execution at runtime. In this paper, we proposed an adaptive redundant reservation strategy, which uses overlapping technique to implement reservation admission and enable resource providers dynamically determine the redundant degree at runtime. By overlapping a new reservation with an existing one, a request whose reservation requirements can not be satisfied in traditional way might be accepted. Also, by dynamically determining the redundant degree, our strategy can obtain optimal tradeoff between performance and reliability for distributed high-performance computing systems. Experimental results show that the strategy can bring about remarkably higher resource utilization and lower rejection rate when using redundant reservation service at the price of a slightly increasing of reservation violations.

The huge amount of biological information implies a great challenge for data analysis, particularly for combinatorial methods such as Multifactor Dimensionality Reduction. This method can be computationally intensive, especially when more than ten polymorphisms need to be evaluated. The Grid is a promising architecture for genomics problems providing high computing capabilities. In this paper, we describe a framework for supporting the MDR method on Grid environments. This framework helps biologists to automate the execution of multiple tests of gene-gene interactions detection. To evaluate the eciency of the proposed framework, we conduct experiments on the Grid5000. A Grid infrastructure distributed in nine sites around France, for research in large-scale parallel and distributed systems. compute-intensive

Virtualized Computing Cloud (VCC) is a well-known resource-provisioning and computing environment due to advantages such as maximized resource utilization, isolated performance, and customizable runtimes. Therefore, recently, VCC has been increasingly adopted in a broad spectrum of service domains, including internet-based application services and computational science. However, efficient resource (i.e., virtual machine; VM) allocation techniques are required to deliver higher-quality Internet and/or computing services. In this paper, we propose a novel dynamic, self-adaptive VMs allocation technique considering network resource contention in a Xen virtualization environment. In addition, we provide generic virtualized resources and job management framework to realize our proposed VM allocation technique. Using the virtualized resources and job management framework, we also practically analyze the impact of various system parameters and job characteristics on the performance of our technique. The results show that our approach outperforms others, reducing the average job response (by up to 37%) and execution (by up to 22.3%) times.

리눅스 Cgroups은 컨테이너 기반 클라우드 서비스 구축에서 각 컨테이너 별 시스템 자원을 할당하기 위한 핵심 적인 역할을 담당하고 있다. 특히 입출력 자원의 경우 리눅스 Cgroups은 컨테이너의 가중치에 따라 입출력 대역폭을 분배하는 기법을 지원하고 있다. 그러나 성능 분석 결과에 따르면 현재 리눅스 Cgroups의 입출력 대역폭 분배 기법은 NVMe SSD와 같은 고성능 저장장치를 사용할 경우 입출력 성능이 크게 저하된다는 한계점을 가지고 있다. 따라서 본 논문에서는 리눅스 Cgroups을 위한 새로운 피드백 기반의 동적 대역폭 분배 기법을 제안하고자 한다. 제안하는 기법은 가중치에 따라 입출력 크레딧을 분배하며 고성능 저장장치의 성능 변화를 동적으로 반영해 입출력 크레딧을 계산함으로 써 저장장치의 성능 저하를 최소화한다. 제안된 기법은 리눅스 커널 5.3에 구현되었으며 성능 평가 결과 정확한 입출력 대역폭 분배를 수행할 뿐만 아니라 기존 기법에 비해 최대 2배 높은 입출력 성능을 보여주었다.

Linux Cgroups takes a fundamental role for sharing system resources among multiple containers on container-based cloud computing environment. Especially for I/O resource, Linux Cgroups supports a mechanism for sharing I/O bandwidth in proportion to I/O weight. However, the current mechanism of Linux Cgroups using BFQ I/O scheduler seriously degrades the I/O performance with high bandwidth storage device such as NVMe SSDs. In this paper, we proposed a new feedback based I/O bandwidth sharing scheme for Linux Cgroups which allocates I/O credits to containers according to I/O weights and adjusts the amount of credits to performance fluctuation of NVMe SSDs. The proposed scheme is implemented on Linux kernel 5.3 and evaluated. The evaluation results show that it can share the I/O bandwidth among multiple containers proportionally to I/O weights while improving I/O performance more than twice as high as the existing scheme.

본 연구의 목적은 두 가지이다. 첫째, 제조업 혁신과 관련하여 이론적인 측면에서 의미, 파급효과, 고려요소 등에 대해 고찰한다. 둘째, 고성능컴퓨팅(HPC) 활용 정책의 위상을 검증하고 미국과 한국의 상황을 분석한다. 각국의 제조업 혁신 정책은 공통적으로 생산성의 획기적 향상을 목표로 하는데 단순한 생산성의 개선이 아니라 패러다임의 전환을 수반하는 혁신 지향 정책으로서의 성격을 갖는다. 장기적인 성장과 고용을 위해서는 탈공업화를 대체하여 재공업화의 필요성도 있다고 보아야 한다. 제조업 혁신을 통해 고용이 일시적으로 또는 부분적으로 줄어들 수 있으나 간접적인 경로로 고용이 확대되는 효과가 더 클 것이다. HPC 활용의 정책은 제조업 혁신의 부분집합으로서가 아니라 별개의 흐름으로서 중요성을 갖는다. 미국의 경우 HPC 기반의 M&S 활동을 촉진하기 위해 정부주도로 애로요인 해소에 주력하고 있고 민관 합동체제를 통한 문제해결 방식을 추진하고 있다. 한국의 경우 HPC 기반의 M&S 활동에 관련된 생태계 조성이 필요하고 이를 위해 제조기업 M&S 활용 확대와 M&S 지원 전문기업 육성의 과제가 중요하다고 할 수 있다.

The purpose of this study is two fold. First, we will explore the meaning, spread effect and consideration factors of manufacturing innovation in terms of theoretical perspective. Second, we will verify the status of high performance computing (HPC) utilization policy, and analyze the situation of US and Korea. Manufacturing innovation policy in each country has the objective in common which aims epoch-making enhancing of productivity. Nevertheless it can be characterized as innovation oriented policy rather than simple trial of productivity improvement. For long term growth and employment, the need for reindustrialization instead of deindustrialization should be recognized. Employment may be decreased temporarily and partially due to manufacturing innovation. However net effect of employment increasing will be bigger because of indirect employment. HPC utilization policy has the importance as a separate movement other than as a subset of manufacturing innovation. US government is trying to eliminate the bottleneck elements in adoption of HPC based M&S activity, and to promote the way of problem solving through the mechanism of public-private partnership, in spite of low level of HPC based M&S. In Korea, ecosystem related with the activity of HPC based M&S is needed, and expansion of M&S utilization in manufacturing companies and fostering of M&S supporting institutions will be important for this task.

Nowadays, Genomic data constitutes one of the fastest growing datasets in the world. As of 2025, it is supposed to become the fourth largest source of Big Data, and thus mandating adequate high-performance computing (HPC) platform for processing. With the latest unprecedented and unpredictable mutations in severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the research community is in crucial need for ICT tools to process SARS-CoV-2 RNA data, e.g., by classifying it (i.e., clustering) and thus assisting in tracking virus mutations and predict future ones. In this paper, we are presenting an HPC-based SARS-CoV-2 RNAs clustering tool. We are adopting a data science approach, from data collection, through analysis, to visualization. In the analysis step, we present how our clustering approach leverages on HPC and the longest common subsequence (LCS) algorithm. The approach uses the Hadoop MapReduce programming paradigm and adapts the LCS algorithm in order to efficiently compute the length of the LCS for each pair of SARS-CoV-2 RNA sequences. The latter are extracted from the U.S. National Center for Biotechnology Information (NCBI) Virus repository. The computed LCS lengths are used to measure the dissimilarities between RNA sequences in order to work out existing clusters. In addition to that, we present a comparative study of the LCS algorithm performance based on variable workloads and different numbers of Hadoop worker nodes.

The last decades have witnessed an increasingly indispensible role of high performance computing (HPC) in science, business and financial sectors, as well as military and national security areas. To introduce key aspects of HPC to a broader community, an HPC session was organized for the first time ever for the United States and Korea Conference (UKC) during 2012. This paper summarizes four invited talks that each covers scientific HPC applications, large-scale parallel file systems, administration/maintenance of supercomputers, and green technology towards building power efficient supercomputers of the next generation.

HPCC (High Performance Computing Class)는 Fortran90으로 구현된 객체지향형 클래스로서 FDM (Finite Difference Method) 모델 프로그램을 메시지 전송 방식을 사용하는 병렬 프로그램으로의 구현을 지원한다. HPCC는 병렬프로그램에서 메시지 전송 방식의 사용을 단순화하고, HPCC에 의해 정의된 객체는 실행 시간에 병렬처리에 관련된 데이터와 관련 메소드를 제공한다. 또한 HPCC는 순차프로그램의 인덱스 전환을 위하여 매크로를 제공한다. HPCC를 사용하여 병렬화를 할 경우에는 병렬화 기간이 현격히 단축될 수 있으며, 메시지 전송을 직접 사용하는 것보다 효율적인 성능을 보인다. HPCC의 유용성과 성능을 알아보기 위하여 Fortran으로 구현된 3개의 레거시 모델 코드를 병렬화하였으며 PC 클러스터를 사용하여 성능의 효율성을 알아보았다.

High Performance Computing Classes (HPCC) is a group of portable object-oriented classes designed with following objectives: (1) to simply the parallelizing process of the Fortran programs based on the FDM (Finite Difference Method) model and (2) to enhance performance of the parallel programs. The objects defined by HPCC provide the parallel programs with the information and the methods required for parallel processing at run-time. HPCC also provides the macros for the index transformation of the sequential programs. To show usefulness and efficiency of HPCC, three different legacy FDM models were parallelized with HPCC, and performance results were presented on a cluster of PC's.

CFD has been used in aircraft development and broaden its influence in various fields of industries. This paper briefly introduces the historical trends of computing system, the overview of CFD applications in Korean Supersonic Trainer Development Program and the demand for CFD software in industry points of view.