High performance distributed data reduction and analysis with the netCDF Operators (NCO).pdfVIP

下载本文档

13
0
约6.45千字
约 13页
2017-04-13 发布于江苏
举报
版权申诉

High performance distributed data reduction and analysis with the netCDF Operators (NCO).pdf

1、有哪些信誉好的足球投注网站（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。。
2、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。
3、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)。
4、该文档为VIP文档，如果想要下载，成为VIP会员后，下载免费。
5、成为VIP后，下载本文档将扣除1次下载权益。下载后，不支持退款、换文档。如有疑问请联系我们。
6、成为VIP后，您将拥有八大权益，权益包括：VIP文档下载权益、阅读免打扰、文档格式转换、高级专利检索、专属身份标志、高级客服、多端互通、版权登记。
7、VIP文档为合作方或网友上传，每下载1次，网站将根据用户上传文档的质量评分、类型等，对文档贡献者给予高额补贴、流量扶持。如果你也想贡献VIP文档。上传文档

High performance distributed data reduction and analysis with the netCDF Operators (NCO)

High performance distributed data reduction and analysis with the netCDF Operators (NCO) Charlie Zender1 and Daniel Wang2 1Department of Earth System Science 2Department of Electrical Engineering and Computer Science University of California, Irvine Thanks to: H. Butowsky (UCI), S. Jenks (UCI), H. Mangalam (UCI), R. Peterson (U. Alaska), R. Rew (Unidata) Presented to: American Meteorological Society 2007 Annual Meeting, San Antonio TX (Web: /smn/smn nco ams 200701.pdf) 1. IPCC Data Reduction Prototype Problem Figure 1: Predicted Global (left) and California (right) annual-mean temperature from 2000–2099 under SRESA1B 720 ppm CO2 stabilization scenario. Temperature scales differ. Our “Holy Grail” is to make Distributed Data Reduction and Analysis (DDRA) more practical, less painful, and quicker for non-computer scientists 2. Three Levels of Optimization To achieve high performance in data processing, NCO identifies and exploits parallelism on multiple-levels, and uses smart arithmetic algorithms: ? Parallelize across commands within script: – Script Workflow Analysis for MultiProcessing (SWAMP) – Schedule execution based on dependency tree analysis – Server-side processing reduces bandwidth for distributed data – Store intermediate files in RAM and transfer results only ? Parallelize arithmetic across variables within file: – Symmetric Multi-Processing (OpenMP) – Message Passing Interface (MPI) ? Optimize low-level, repetitive arithmetic algorithms: – Most Rapidly Varying (MRV) – Weight Re-use (WRU) Table 1: NCO Operator Summary (Zender , 2006) Command Name (primary functionality) Type MFO Par. ncap Arithmetic Processor (algebra, scripts) A ncatted Attribute Editor (change attributes) M ncbo Binary Operator (subtraction, addition) A X ncea Ensemble Averager (means, min/max) A X X ncecat Ensemble Concatenator (join files) M X X ncflint File Interpolator A X ncks Kitchen Sink (sub-set, hyperslab) M ncpdq Pack Data, Permute Dimensions A/M X ncra Record Averager (means,