The New Linux ’perf’ tools(新的Linux性能的工具).pdf

The New Linux ’perf’ tools(新的Linux性能的工具).pdf

  1. 1、本文档共42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
The New Linux ’perf’ tools(新的Linux性能的工具)

The New Linux ’perf’ tools Linux Kongress September, 2010 Arnaldo Carvalho de Melo acme@ Presentation Flow . Motivation . Focus on the tools . But some kernel details will be mentioned . Some demos hopefully at the end How did I get involved? . I am no specialist on performance counters . pahole the dwarves . ELF, DWARF, symtabs, dynsyms, relocations, etc . ftrace How did I get involved? Part two . Part of the Red Hat Real Time team . We need to discover why 100us deadlines are not being met . Why this is slower on your RT kernel than on the RHEL one? . Observability tooling! . Huge educational value, use it! Renewed interest in profiling tools . Complexity of systems growing . Pervasiveness of multithreading . Hardware assists Performance Counters . Performance counters are special hardware registers . Available on most modern CPUs . Count the number of some hw events instructions executed cache-misses suffered branches mispredicted . Without slowing down the kernel or applications . Can trigger interrupts when a number of events have passed Limited resource: Some are programmable, some are for specific events. Processor: UltraSparc 2 Pentium III 2 Athlon 4 IA-64 4 POWER4 8 Pentium IV 18 Nehalem 7 Tracepoints Static probe points that are put in place by subsystem maintainers and that can be enabled later. Dynamic probe points Dynamicly inserted probe points using hardware breakpoints. The oprofile development problem . Disconnected kernel userspace development . Linus problem with Atom and Nehalem support . Less of the 2 broken pieces approach - one working piece . /Articles/339406/ The perf user interface approach . git like . Many subcommands . Per thread/per workload/per CPU

您可能关注的文档

文档评论(0)

jiupshaieuk12 + 关注
实名认证
内容提供者

该用户很懒,什么也没介绍

版权声明书
用户编号:6212135231000003

1亿VIP精品文档

相关文档