초고성능컴퓨팅 계산과학공학 연구 및 선도기술 개발

Abstract

ㅇ 초고성능컴퓨팅 활용 계산과학공학 선도 연구 - 초고성능 컴퓨팅 활용 전 속도 영역 유동해석 기술 및 활용 연구 - 전 속도 영역 유동해석 SW 개발 - 유동해석 SW 정밀도 향상을 위한 핵심 기술 개발 ㅇ 거대 컴퓨팅 가시화 기술 개발 - 세계 수준의 거대 컴퓨팅 계산과학공학 가시화 기술 개발로 R&D 효율화 및 선도 기술 확보 - 기술의 실용화 및 기술이전으로 거대 컴퓨팅 가시화 기술 활용확대 ㅇ 데이터 기반 계산과학공학 플랫폼 구축 선도 연구 - 계산과학공학 플랫폼 성능 개선을 위해 기존 결과를 저장하여 활용할 수 있는 선도 기술 연구 - 사용자가 요청한 시뮬레이션을 보다 적은 비용으로 처리할 수 있는 데이터 주도형 시뮬레이션 서비스 기술 개발 - 기존 연구와는 다른 이전 시뮬레이션 결과를 추후 요청되는 시뮬레이션의 성능을 높이기 위해 재활용 - 존재하지 않는 시뮬레이션에 대한 통계적 기계학습 기법을 적용하여 결과를 예측 - 예측 및 검색 서비스를 통해 플랫폼 성능 향상 및 보급을 통한 커뮤니티 육성 지원 ㅇ 플랫폼 보급을 통한 계산과학공학 커뮤니티 활성화 - 국내외 관련 기술 워크숍 및 성과 교류 활동 수행 - 계산과학공학 플랫폼 여름학교 개최 및 인력 교육 - KSC, ISC, SC 연계의 커뮤니티 체계화 및 육성

IV. Results of the study o Studies on computational science using supercomputer - A study on the flow analysis and application using high performance computing in all speed region · CFD SW development for flow analysis of all speed regions · Development of three dimensional laminar-turbulent transition model · Development of numerical model for unsteady separation phenomenon analysis · CFD SW verification using standard problems

o The massive data visualization technology development - The performance improvement for massive data visualization · The transfer time for visualization results has cut off 51% by improving network communication structure · The real-time visualization performance for 3 TB data has reached to 7.35 seconds · The massive data loading speed has been improved to 2 times faster than before - The low-cost optimal visualization technology development for popularization · The development of 3 core technology for performance improvement · The visualization user response performance of 1 TB data has reached to 2.4~4.3 times faster than ParaView · The amount of memory usage for unstructured grid volume rendering algorithm has been decreased to 49% - The leading technology development for technological innovation · The HD visualization streaming encoding/decoding performance has reached to more than 59 fps when more than 3 people uses the system with more than 2 kinds of devices · The development of information-theory based auto-detection technique for turbulence collapse point

o Research on advanced technology for data-based computational science & engineering platform - R&D on data-driven simulation service system technology · Simulation data translator-loader: extracting the information of input and output variables and their respective values from simulation results and convert such information into a JSON document and loading the document into NoSQL databases · Simulation databases: managing simulation data via MongoDB · Simulation query interface: providing an interface of retrieving and returning results from simulation databases and reporting predicted results · Simulation result predictor: considering a variety of statistical machine-learning models and providing predicted results with high accuracy and low error rate ▹ Reporting the predicted results of a specific method with the lowest error rate after applying ten different kinds of statistical machine learning methods ▹ Showing a performance gain of about 3000x speedup in simulation execution: an average of about fifty-minute execution time vs. an average of less than one-second prediction time ▹ Showing an average error rate of less than 3% on airfoil simulations: 0.9% to 7% · Metadata databases: supporting multiple simulation programs - Activation of computational science & engineering community through platform dissemination · Application and distribution of computational science & engineering platform domestically and abroad ▹ Application of data-based computational science & engineering platform ▹ Grand prize awarded in Internet Eco-Awards Social Contribution Innovation ▹ Cooperated with Taiwan NCHC (National Super Computing Center), NCSA (National Super Computing Center) and Vietnam ICST (Computational Science Engineering Institute) · Performing related technology workshops and performance exchange activities at home and abroad ▹ Proliferation of performance through participation in ISC, SC, KSC ▹ Hold workshop for spreading computational science engineering simulation platform · Calculation Scientific Engineering SW development Human resource training and SW development ▹ Training of computational science & engineering software developers (47 persons) to achieve R&D efficiency through computational science ▹ Development of computational science & engineering software (total 47) and system installation service