GOCI-II 자료처리를 위한 병렬 알고리즘의 기술적 분석

Abstract

Through the application of parallel programming languages such as Open Specification of Multi-Processing (OpenMP), Open Computing Language (OpenCL) and Message Passing Interface (MPI) to Geostationary Ocean Color Imager (GOCI) ocean color algorithms, we found the most appropriate parallel processing methodology for successful GOCI-II development. We tested algorithms of Total Suspended Solid concentration (TSS), Chlorophyll concentration (chl) and absorption coefficient of Colored Dissolved Organic Matter (CDOM) using aforementioned parallelism techniques. In experimental results from Chundoong cluster at Seoul National University, speedup was 2.56, 1.40, 1.73 in TSS algorithm, 3.92, 2.56, 2.67 in chl algorithm, and 4.45, 2.62, 2.96 in CDOM algorithm, respectively. Similarly, efficiency was 1.28, 0.70, 0.17 in TSS algorithm, 1. 96, 1.28, 0.17 in chl algorithm and 2.23, 1.31, 0.16 in CDOM algorithm, respectively. Also, we could evaluate the degree of parallelism level of each algorithm using Amdahl’s law and Karp-Flatt metric. Consequently, experimental results came close to the ideal speedup and it showed the higher efficiency in the OpenMP.lgorithms, we found the most appropriate parallel processing methodology for successful GOCI-II development. We tested algorithms of Total Suspended Solid concentration (TSS), Chlorophyll concentration (chl) and absorption coefficient of Colored Dissolved Organic Matter (CDOM) using aforementioned parallelism techniques. In experimental results from Chundoong cluster at Seoul National University, speedup was 2.56, 1.40, 1.73 in TSS algorithm, 3.92, 2.56, 2.67 in chl algorithm, and 4.45, 2.62, 2.96 in CDOM algorithm, respectively. Similarly, efficiency was 1.28, 0.70, 0.17 in TSS algorithm, 1. 96, 1.28, 0.17 in chl algorithm and 2.23, 1.31, 0.16 in CDOM algorithm, respectively. Also, we could evaluate the degree of parallelism level of each algorithm using Amdahl’s law and Karp-Flatt metric. Consequently, experimental results came close to the ideal speedup and it showed the higher efficiency in the OpenMP.