grapes有限区域切线伴随模式高效并行算法研究-research on efficient parallel algorithm of tangent adjoint mode in grapes finite area.docx

下载文档 降价啦

0
0
约4.56万字
约 125页
2018-05-29 发布于上海
举报
版权申诉
保障服务

grapes有限区域切线伴随模式高效并行算法研究-research on efficient parallel algorithm of tangent adjoint mode in grapes finite area.docx

1、本文档共125页，可阅读全部内容。
2、原创力文档（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。
3、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击这里二次下载。
4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。

grapes有限区域切线伴随模式高效并行算法研究-research on efficient parallel algorithm of tangent adjoint mode in grapes finite area

????????????????乘??????П????????????????????????????????????催???乘????????????????????????????????????????????????????????????????????????乘???GRAPES?Global/RegionalAssimilationandPredictionSystem???????????GRAPES-4DVAR?г?????????????????????????GRAPES???????????????????催?????????????????お????????ゴ???????????????????????ㄝ????催??????????????お???????????????????????????1??GRAPES?????????????????お?催?????⑤???????????????????????Ё???????????乜????????????????????????催?25%??2????????????????ü?????????????30%??????????????100%?????3????????????????????????????????????????-???????4???GRAPES?????????????????????乬?????????????????????ぎ??????Ё????????????????????????????????????????128?????70%????????5???GRAPES???????????????????????????????OPENMP???????????MPI??????????????MPI???GRAPES??????????MPI???????90%???????????????催5%?10%??????????GRAPES????????????????????ABSTRACTFour-dimensionalvariationalassimilationasoneofthekeytechnologiesofQXPHULFDOZHDWKHUSUHGLFWLRQ?scantaketheinformationrelatedintimeforobserveddataintoaccounttoimprovethequalityofinitdatawhichdeterminetheeffectofforecast.Itcanbeassimilatedthedifferenttimes,differentregions,differenttypesofobservationaldatabeconsideredthemosteffectiveschemeinternationalindataassimilationcurrently.Butitscalculationisverycomplicatedandneedsmorecomputationsandmoretimetocompute.Thefour-dimensionalvariationalassimilationsystemofGRAPES(Global/RegionalAssimilationandPredictionSystem)calledGRAPES-4DVARforshortwhichisanewgenerationofnumericalweatherpredictionsystembedevelopedbyChineseindependentlyhavethesimilarfeaturewithalargeamountofcomputations,needingmorememoryandlongertimewhenrunning.Howtoreducetheelapsedtimebyimprovingthecodeefficiency,changingthealgorithm,enhancingtheparallelscalabilityisthekeyandfocusofthisarticle.Thisarticlemainlyfocusonhowtoobtaintheperformancefromoptimizedcodeforimprovingefficiency,howtoanalysistheimpactonprogramperformancebyusingadifferentwaythroughthequantitativemethod,andhowtouseamixedparallelmodeforincreasesca