목차

1. 서론 ························································································ 1
2. 사전 연구 ················································································· 4
2.1 프로세서 중심 컴퓨터 구조 ····················································· 4
2.2 메모리 중심 컴퓨터 구조 ························································ 6
2.3 주요 MPI 함수 및 통신 오버헤드 ············································· 8
2.3.1 MPI_Send 및 MPI_Recv ··················································· 8
2.3.2 MPI_Bcast ······································································· 8
2.3.3 MPI_Reduce ···································································· 9
2.3.4 MPI_Scatter ···································································· 9
2.3.5 MPI_Gather ···································································· 9
2.4 Docker를 이용한 분산 병렬 시스템 구축 ································ 11
2.4.1 Docker 이미지 Ubuntu 다운로드 ······································ 11
2.4.2 Docker 컨테이너 hpc-mpi 생성 및 설정 ····························· 12
2.4.3 Docker 이미지 hpc-mpi 생성 ··········································· 16
2.4.4 Docker 컨테이너 hpc-mpi-100~115 생성 ························ 17
2.4.5 MPI 병렬 프로그램 컴파일 및 수행 ····································· 18
3. MC-MPI-LIB 설계 및 구현 ························································ 20
3.1 MC_MESSAGE 형 ······························································· 21
3.2 도우미 함수들 ······································································· 22
3.3 메모리 중심 컴퓨터 구조를 위한 수정된 MPI 함수 ···················· 23
3.4 기존 MPI 함수와 메모리 중심 컴퓨터 구조를 위한 수정된 MPI 함수
의 통신 오버헤드 비교 ·································································· 34
3.4.1 MPI_Send & MPI_Recv ··················································· 34
3.4.2 MPI_Bcast & MPI_Reduce ·············································· 34
3.4.3 MPI_Scatter & MPI_Gather ············································ 35
4. MC-MPI-SIM 설계 및 구현 ······················································ 36
4.1 성능 측정 방법 ····································································· 38
4.1.1 MPI 병렬 프로그램 전체 수행 시간 측정 ····························· 40
4.1.2 PC용 MPI 병렬 프로그램 통신 부분의 수행 시간 측정 ·········· 41
4.1.3 MC용 MPI 병렬 프로그램 통신 부분의 수행 시간 측정 ········· 43
4.1.4 MPI 병렬 프로그램 계산 부분의 수행 시간 측정 ··················· 45
4.2 성능 측정 데이터 저장을 위한 코드 삽입 ································· 46
5. 성능 측정 ·················································································49
5.1 성능 측정 환경 ······································································ 49
5.2 성능 측정에 사용한 MPI 병렬 프로그램 ··································· 50
5.3 성능 측정 결과 ······································································ 51
5.3.1 Ping-pong ······································································· 51
5.3.2 Pi-Calculation ·································································· 54
5.3.3 Matrix-vector multiplication ············································ 56
5.3.4 Sorting ············································································ 58
5.3.5 All-pairs shortest paths ··················································· 60
6. 결론 ························································································ 62
참고문헌 ····················································································· 64
ABSTRACT ················································································· 67