Please note that User Registration has been temporarily disabled due to a recent increase in automated registrations. If anyone needs an account, please request one here: RequestAccount. Thanks for your patience!--Wmat (talk)
Please email User:Wmat if you experience any issues with the Request Account form.

Difference between revisions of "ITJ2005Detail 1-5"

Jump to: navigation, search
m (Bot (Edward's framework))
Line 24: Line 24:
* [[Media:KFI-presentation.ppt]]
* [[Media:KFI-presentation.ppt]]
* [[Media:omap-serial_init.trace.txt]]
* [[Media:omap-serial_init.trace.txt]]

Latest revision as of 09:56, 14 July 2007

Technical Discussion: "Learning the Kernel, and Finding Performance Problems, with KFI"

by Tim Bird (Sony)


Kernel Function Instrumentation is a simple tracing system for the Linux kernel, which is easy to use and very powerful. Profiling systems, such as OPROFILE, perform a statistical sampling of execution locations over a period of time. Other event tracing systems, like LKST or LTT, utilize a relatively small number of fixed tracepoints to analyze state transitions and event flow in a Linux system. KFI differs from these in that it provides highly detailed and comprehensive call-graphs with timing data (essentially instrumenting EVERY function in the kernel).

Because of this high level of detail, KFI is useful for learning how the kernel operates, and because of KFI's runtime filtering and post-processing tools, it is especially useful for finding long delays and performance problems in the Linux kernel. This tool has been used extensively by the Bootup Time Working Group of CELF to find long delays in kernel startup.

This talk give an introduction to KFI and give a tutorial on its basic use. Also, I will present some new features of KFI that have just been released, including support for dynamic traces (ability to configure and start a trace from user space via a /proc interface), and a new tool for showing threaded call graphs.

Presentation materials