- 1、本文档共35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)。
查看更多
HDFS简介 杨林三 2012.08.28 提纲 文件系统 HDFS 实验 文件系统概览 文件系统数据结构 文件系统数据结构 文件系统工作流程 Linux Ext2 Linux Ext3 NFS 小结 Hadoop Hadoop HDFS目标 Targets Characters Very large files GB, TB, PB+ Millions of files 10K+ nodes Streaming data access Write-once, read-many-times Good data coherency Time to read whole dataset matters more than time to seek Commodity hardware Cheap Node failure: the chance over large cluster is high Redundancy through replicas Failure handling and recovery HDFS目标 Targets Characters Optimized for batch processing move computation rather than data locations of data exposed to computation Not good for Low-latency data access ms High throughput against Low-latency Lots of small files Metadata may run out of memory Seek time over read time Multiple writers, arbitrary file modification A single writer Write at end Write-once, read-many-times HDFS架构 HDFS架构 Data Storage Blocks/Chunks Block size 64/128/256MB Configurable why so big? transfer over seek seek:10ms, transfer rate: 100MB, seek time/transfer time = 1% = block size = 100MB blocks of a file distibuted over datanodes as possible a single FILE, existing of N blocks, is stored on M nodes Namenode Secondary Namenode Why? Peroidically Merge edits and fsimage Prevent edit log from being too large Provide checkpoint of Namenodes fsimage as backup in case of Namenode crash Namenode High-Availability FsImage and EditLog are central data structures of HDFS. A corruption of these files can cause a HDFS instance to be non-functional. For this reason, a Namenode can be configured to maintain multiple copies of the FsImage and EditLog. Multiple copies of the FsImage and EditLog files are updated synchronously. Meta-data is not data-intensive. Namenode Safemode Merge FSImage and Edits Enter safemode Offers read-only view of filesystem to clients Block-location information collection block-location info not in fsimage datanodes check in to report their blocks check blocks minimal replication condition: 99.9% blocks meet min
文档评论(0)