국가 슈퍼컴퓨팅 서비스 개발 및 기술연구

Abstract

□ 사업의 목표: 세계 수준의 슈퍼컴퓨팅 R&D․시범서비스 도출 및 독자 연구개발 역량 확보 □ 주요사업 내용 ○ 엑사스케일을 위한 시스템 기술 및 활용 연구 - 국가차원 HTC 서비스 기술 완성 및 글로벌 인프라 연동 활용 확대 - 엑사스케일을 위한 슈퍼컴퓨팅 핵심기술: 고성능 I/O 처리기술 선행연구 수행 ○ 고성능 컴퓨팅 클라우드 기술 개발 - 컴퓨팅 클라우드 관리 및 클라우드 인프라 연동 통합 작업관리 기술 개발 - 고성능 병렬처리를 위한 분산 스토리지 시스템 기술 개발 ○ 한국형 신종 감염병 조기 대응 시스템 기술 개발 - 독감바이러스 원인체 규명을 위한 통합분석시스템 원형개발 - 감염병 확산 시뮬레이션을 위한 공공데이터 활용연구 ○ 비주얼 슈퍼컴퓨팅 기술개발 및 적용 - 대용량 데이터 가시화 기술 개발 - 비주얼슈퍼컴퓨팅 기술 프레임 워크 설계 및 개발 ○ 국가 슈퍼컴퓨팅 전문인력 양성 및 교육훈련 - 국가 차원의 교육 추진체계 마련 및 슈퍼컴퓨팅 활용인력 양성, 교재 개발 - 선도인력 양성용 응용연구 ○ 슈퍼컴퓨팅 활용기술 연구 - 멀티스케일 모델링 개발 및 슈퍼컴퓨팅 활용연구

Ⅳ. Result of the study ㅇ Development of system technology for exascale and its applications - Development of Many Task Computing Technology • Development of HTCaaS scout for profiling grid resource • Development of Federated Cloud - Development and application of exascale system technology • Deduce the direction of further research • Develop the large amount data transfer method by international collaboration and apply it to the tokamak plasma analysis research. - High performance distributed computing applications • Improve HTCaaS based on PLSI • Support its applications including Bio insilico experiment, HEP and hadron collider experiments. • Improve metadata catalogue AMGA and user support. • Develop Bell II data handling system • Geant4 R&D and application research - Research based on supercomputing • Development of core modules in LAMMPS-based multi-scale modeling (algorithm for the boundary communications in MD+SRD) ㅇ Research and Development of High Performance Computing Cloud Technologies - Development of resources and jobs management technologies which incorporate virtual machine based computing cloud resources • Architectural design and implementation of advanced heterogeneous computing resources and jobs management framework • Research and development of technologies which minimize virtual machine image creation time and increase the performance of file I/O on virtual machines • Research and development of technologies which allocate and schedule both virtual machines and simulation jobs dynamically considering network resource contention • Development of Web standard based RESTful OpenAPIs for Cluster-On-Demand, Cluster-Extension, and Cloud Auto-Scaling services • Development of cloud service scenarios on requests and SCLOUD (Scalable-Cloud) technology • Design and implementation of general SCLOUD elastic resource driver which incorporates onto various physical cluster and application environments (e.g., SLURM, HTCondor, Hadoop, etc.) • Development of SCLOUD CLI tools - Development of a distributed storage system for high performance parallel applications • Development of high performance meta-data lookup technology ∎ Proof-of-concept development of searchable file system using file extended attributes ∎ Architectural design for rapid processing using auto-shaded large-scale meta-data ∎ Development of fast and powerful meta-data Search/Management APIs • Development of Metadata Importer for automating file provenance management ∎ Development of application-specific plug-in modules for user-defined meta-data extractions ∎ Development of application-specific plug-in auto-execution modules ∎ Development of large-scale data management framework based on high-performance meta-data lookup technology ㅇ Application of bulk public informations for developing simulation protocol of infectious disease spreading in Korea - High-performance, high-capacity computing resources based on user requirements linked to influenza genome analysis automation platform development (domestic patent 2, program registration 2, SCI journal publication) - Construction database on integrated public data for epidemic simulation (http://lcbb.snu.ac.kr/kodes) - Supercomputer based international organization, the Korea-US joint research team to build the analytical system and perform a full analysis research in cancer (SCI journal publication) - Host 2 education courses, and 6 media announcement. ㅇ Development and application of the visual super computing technology - Development of large data visualization technology • Achieved 1.9 times better suability and 48% higher performance than the only commercial visualization SW supporting distributed/parallel visualization in the thermal fluid field • Achieved 16 times better stability compared with one of the world class visualization tools, VisIt • Achieved from 2.28 to 45.9 times higher performance than Ensight (a commercial visualization SW) • For practical use, the developed system was released on the EDISON CFD community • Our visualization system won the 2nd place (i.e. golden award) from the international military technology contest and it demonstrates the outstanding features and benefits of the results of the study - Design and development of visual super computing technology frame work • Designed technical framework for the target application that was figured out from the survey of technical needs on visual super computing • Developed the steering technology that can start a simulation, stop the simulation, and restart the simulation in the unified visual super computing environment • Achieved 70% automation rate for the grid construction in pre-processing • Achieved 30% performance improvement on interpretation by porting a part of CPU code to GPU code • With the optimization for parallel post-processing visualization, it shows less than 2 seconds response time and achieved more than 40 frames/sec rendering performance