사용자 환경 개선을 위한 초고성능컴퓨팅 플랫폼 개발

Abstract

□ 사업의 목표: 국가 슈퍼컴퓨팅 인프라의 효율적 활용을 위한 서비스 플랫폼 및 차세대 슈퍼 컴퓨팅 시스템 소프트웨어 개발을 통한 슈퍼컴퓨팅 세계 7대 강국 실현에 기여 □ 주요사업 내용 ○ 서비스 플랫폼 기반 기술 개발 및 구축 - 응용에 특화된 멀티 서비스 체계 설계 - 플랫폼 프레임워크 구축 - 응용적용 및 커뮤니티 활성화 ○ 운영 관리 SW 설계 및 개발 - 슈퍼컴퓨터 장애 예측 기술 선행 연구 및 개발 - Belle II 실험 지원을 위한 AMGA 개발 및 유지보수 ○ 슈퍼컴 시작품 설계 및 고성능 I/O 기술 개발 - 데이터 집약형 슈퍼컴퓨터 요구분석 및 설계 - 고성능 I/O 기술 개발

Ⅳ. Result of the study ㅇ Development of Service Platform - Analysis of multi frameworks for various applications ● Test bed for analysis of MTC/HTC framework ● Monte-Carlo, Autodock, vina test - Development of service platform framework ● Apply the Argonne National Lab’s Swift to HTCaaS ● Design a HTC cluster management system ● 21% erformance improvement with new job scheduler ● HTCaaS user data encryption ● Join in EGI Federated Cloud ● Service for various applications ● Virtual screening with KRIBB-Sangju univ using 617 CPU cores ● Monte-Carlo simulation for New Physics two induced sci papers ㅇ Design and Development of Supercomputer management SW - Preceding research and Development of supercomputer failuer prediction technology ● Collecting demands of department of supercomputing infrastructure and cooperative research and development of vCLAM which is the monitoring and supercomputer log analyzing tool * visualization tool of Cluster Log Analysis and Monitoring ● Research of log collecting and analysis generic technologies such as Apache spark, plum, Kafka for failure prediction and monitoring ● Designing the prototype of supercomputer log analysis framework for real-time monitoring and analyzing huge log data ● Development of a test bed of real-time log analysis system - AMGA Development and maintenance for supporting Belle II experiment ● Releasing the officail version of AMGA 2.5.0 to fix the problem of the response time decreases according to increasement of the number of directories ● Maintenance and support services of AMGA : fixing urgent error and providing patched version ㅇ Designing I/O intensive supercomputer and Research on high performance I/O system - Requirement analysis and design of data-intensive supercomputer ● Design and implementation of high-performance I/O testbed for performance analysis, reflecting the storage architecture of 4th KISTI supercomputer, and deduction of performance insights by performance comparison and analysis with 4th KISTI supercomputer ● Write performance achieved maximum 120% by optimizing communication layer of lustre filesystem ● Write performance achieved maximum 160% by implementation of burst buffer PoC system and technology development - Development of high-performance I/O technologies ● Development of abstract I/O framework enabling asynchronous and transform-based remote data movement ● Reduce 50% of time spent on analysing ECE-Imaging data at remote server