1. 1、本文档共12页,可阅读全部内容。
  2. 2、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。
  3. 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. 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
hic2011-hdfs2

HDFS2, 一种分布式 NN 实现 2011-12-02 孙桂林 Why? Scalability Availability ● 10,000 Nodes ● No Single Point ● 1,000,000,000 Files ● Failover The limits to growth Memory of 1,000,000,000 files ● 380GB (1.1 blocks per file) 1.1 ● 860GB (3 blocks per file) 3 ● 1300GB (5 blocks per file) 5 Performance ● Latency and/or Throughput and/or HDFS Federation? Scale of cluster VS Scale of namespace VS Do we need a larger namespace? ● Scale of computing Locatily ● Easy to use (directory moving) ● Operation cost (data migration) Solution Shared Object Management Layer Light-weight namespace ● Without blocks management ● Without file attributes ● Move infrequently accessed data to disk (optional) Namespace Federation (optional) Shared storage (optional) Architecture Grow Up Online VS Offline VS Data VS Metadata VS Memory Namespace ● 1,000,000,000 files 66GB ● 2,000,000,000 files 132GB ● … Object Management Layer ● 100,000,000 files per node (60~80GB). Performance Latency ● Batch operation (Group by Node) Throughput ● Meta operation ● Internal/External block operation ● Overall throughput increased by 500% ~ 1000% Availability Single point fa

文档评论(0)

牛X文档 + 关注
实名认证
内容提供者

该用户很懒,什么也没介绍

1亿VIP精品文档

相关文档