计算机应用与软件
計算機應用與軟件
계산궤응용여연건
COMPUTER APPLICATIONS AND SOFTWARE
2014年
5期
42-45
,共4页
赵伟%雷晓燕%陈德训%赵春%宋振亚
趙偉%雷曉燕%陳德訓%趙春%宋振亞
조위%뢰효연%진덕훈%조춘%송진아
神威蓝光%海洋环流模式%加速比
神威藍光%海洋環流模式%加速比
신위람광%해양배류모식%가속비
SW supercomputer%Ocean circulation mode%Speedup ratio
基于中国自行研制的超大规模并行计算机神威蓝光平台,对海洋环流模式POP进行移植和进一步优化,并进行模式和机器的并行测试。测试结果表明:0.1度分辨率在5000核以内达到了线性加速比,随着使用线程数目的增多,虽然加速比依然增加,但是加速效率在降低,当线程数目达到24000的时候加速比出现降低,加速效率只有12.6%;0.05度分辨率可在12000核以内达到了线性加速比,同样随着线程数目的增多加速在增多但加速效率在降低,当线程数目达到32000的时候加速比出现降低,加速效率只有25.7%。经过优化后的海洋环流模式POP在神威蓝光并行计算机系统上具有良好的兼容性,并行效率较高,而且具有很强的可扩展性,可用来开展超大规模的并行计算。对测试结果的进一步分析发现,发展并行性良好的线性方程组求解方案、快速高效的计算网络协议等问题是未来发展高分辨率模式中需解决的问题。
基于中國自行研製的超大規模併行計算機神威藍光平檯,對海洋環流模式POP進行移植和進一步優化,併進行模式和機器的併行測試。測試結果錶明:0.1度分辨率在5000覈以內達到瞭線性加速比,隨著使用線程數目的增多,雖然加速比依然增加,但是加速效率在降低,噹線程數目達到24000的時候加速比齣現降低,加速效率隻有12.6%;0.05度分辨率可在12000覈以內達到瞭線性加速比,同樣隨著線程數目的增多加速在增多但加速效率在降低,噹線程數目達到32000的時候加速比齣現降低,加速效率隻有25.7%。經過優化後的海洋環流模式POP在神威藍光併行計算機繫統上具有良好的兼容性,併行效率較高,而且具有很彊的可擴展性,可用來開展超大規模的併行計算。對測試結果的進一步分析髮現,髮展併行性良好的線性方程組求解方案、快速高效的計算網絡協議等問題是未來髮展高分辨率模式中需解決的問題。
기우중국자행연제적초대규모병행계산궤신위람광평태,대해양배류모식POP진행이식화진일보우화,병진행모식화궤기적병행측시。측시결과표명:0.1도분변솔재5000핵이내체도료선성가속비,수착사용선정수목적증다,수연가속비의연증가,단시가속효솔재강저,당선정수목체도24000적시후가속비출현강저,가속효솔지유12.6%;0.05도분변솔가재12000핵이내체도료선성가속비,동양수착선정수목적증다가속재증다단가속효솔재강저,당선정수목체도32000적시후가속비출현강저,가속효솔지유25.7%。경과우화후적해양배류모식POP재신위람광병행계산궤계통상구유량호적겸용성,병행효솔교고,이차구유흔강적가확전성,가용래개전초대규모적병행계산。대측시결과적진일보분석발현,발전병행성량호적선성방정조구해방안、쾌속고효적계산망락협의등문제시미래발전고분변솔모식중수해결적문제。
SW supercomputer,which is a super large scale parallel computer developed solely by China,is ported for ocean circulation mode POP and further optimized.In addition,mode and machine parallel testing is carried out.Test result show s:0.1 degree resolution reaches linear speedup ratio within 5 000 cores.As the number of threads in use increases,speedup ratio still increases but speedup efficiency decreases.When thread number reaches 24 000,speedup ratio starts to decrease while speedup efficiency remains only 12.6%. 0.05 degree resolution reaches linear speedup ratio within 12 000 cores.Similarly as the number of threads increases,speedup increases but speedup efficiency decreases.When thread number reaches 32 000,speedup ratio starts to decrease while speedup efficiency remains only 25 .7%.The optimized ocean circulation mode POP bears fine compatibility on SW supercomputer,higher penalization efficiency and strong expandability,so that it can be applied to very large scale parallel computing.Further analysis on the test result discovers that it is the demanding question in future high-resolution mode development to develop well-parallelized linear equation group resolving solution,quick and efficient computing network protocol as well as other questions.