Linux Kernel State Tracer
Linux Kernel State Tracer(LKST) records information as trace data about
events in the Linux Kernel. It records various events like process context
switch, send signal, exception, memory allocation, send packet, and so on.
User can use LKST for the following.
Analyzing kernel error
LKST can trace the execution pass of kernel and collect the state information at the time.
From these LKST data, user can know where the execution of kernel ended and the kernel status at the time.
Thus user can use LKST as a tool to detect the cause of error.
This function is included in all version of LKST.
The development of this program is partly supported by IPA
(Information-Technology Promotion Agency, Japan).
LKST with lkstlogtools can collect the queue information and the execution time of some key processings and analyze them.
lkstlogtools provides some tools for analyzing and visualizing LKST data too.
From these analyzed data, user can know what processing is slow and the status of the queue at the time.
Thus user can use LKST with lkstlogtools as a tool to evaluate kernel performance.
We released lkstlogtools v1.0.1.
For using it, LKST v2.2.1 is needed.
We will combine this function in the next version of LKST.
LKST is a facility to analyze the problem of Linux Kernel trouble and the performance.
It becomes possible that developers investigate troubles by this facility without
stopping the important(used for Mission Critical purposes) server.
Comparison with Standard Kernel, and LKCD
Functions of present Linux could not offer enough information for developers
to solve the trouble of the system.
||When the kernel is panic, it displays only the registers and stack
information. If developers want more information, it is necessary to
insert printk() in the kernel and recompile it.|
||LKCD stores a dump when the system crashes. But, if developers use it
only, they cannot investigate the sequence of events till kernel panic.
(LKCD is a.k.a. Linux Kernel Crash Dumps)|
Event information of the kernel is always recorded as trace data on the
server under running by using LKST. So, these problems can be solved.
- It is possible to change dynamically which events are recorded. Developers
can obtain information about the events which they concern only. And, It
reduces the overhead of components which is not related with trouble.
- It is possible to change each handlers related each events. A default
handler of all events is is just recording the events. But, if it is
necessary, default handler can be changed. This function can be used as
- Notify user processes when the handler detects abnormal status.
- Change a maskset,which controls what type of events should be recorded,
dynamically by the handler.
-> The system can usually run with a few
events trace for the cause of good performance. And, when the handler
detects abnormal status, it can change a maskset to get more detail
Djprobe - Direct Jump Probe - : Low overhead dynamic probe function
Project Mailing Lists : email@example.com
All Rights Reserved, Copyright (C) 2001, 2005
All Rights Reserved, Copyright
(C) 2001, Fujitsu, Ltd.
This website is hosted by