[Lkst-users] Re: lkstbuf questions?

[Haijin Y. <ya...@cs...>]

>
> Yes, it is only place where OVWRTN_REC appears.
> The LKST_OVWRTN_REC event type means it is an invalid event information.
> The event type is read when following situations.
> (1) Overwrite was occurred during the reading the event, so it cannot be
>  assured that the event is right information.
> (2) Reading empty event information in a buffer.
> You can ignore the events. And you can filter them by using
> "lkstbuf print -e \!overwritten_record" option(see "man lkstbuf").
> Currently, changing maskset cannot filter the event, supporting
> this function is under consideration.
> But appearing 99% of the event in a log file is very strange.
> It is under investigation about this problem.
> Could you tell me more detail about them?
>
The situation happened when my kernel is smp version and seems the two
cpus content on each other for logging under lkstlogd, so wired. right
now, I recompiled the kernel without SMP support, seems better. only the
begging and the ending parts of the trace contains these overwritten_records.

>
> #lkst stop			/* stop lkst */
> #lkstbuf creat -s buffer_size	/* create new buffer has enough size */

What is the max buffer size can you create if RAM is plenty, say, 1G?

> #lkstbuf shift			/* shift to the new buffer */

when I shift to the last buffer, I can not shift back(ioctl error),
Does this mean I can only use the current buffer under "lkst start"
while I can use buffer in a ring manner under "lkstlogd"?
If so, for example, like this, buffer 0 is a waste of mem while I can not
delete it.

 id	 cpu	size	prev_id	next_id	    readp	    writep	wrap
  0	  0     65536	NUL	NUL	     4096	     4480	 0
  1	  0*    65536	NUL	NUL	     4096	      256	 1

root@sunshine: lkstbuf del -b 0
ioctl() error: Invalid argument


> #lkst start			/* start lkst */
> #<command or procedure what you want to trace>

How long can you log the kernel with you max capacity using "lksk"?
I guess I am stupid, if buffer overflow, you just overwrite it and
continue the logging, you are not going to stop when buffer overflow,
right?

> #lkst stop			/* stop lkst */
> #lkstbuf read -f filename	/* read the buffer */
>

>
> "lkst start" starts writing events records into a buffer in the memory.
> "lkstlogd" starts writing the records in the buffer to a file.
> It depends on a user's intention which is used.
> "lkstlogd" causes overhead for writing a file, and "lkst start"
> causes overwriting as said above.
>
If I want to log disk activities, because lkstlogd itself involves lots of
disk writes, the measurement itself is not correctly. I can only log to
memory and dump them to disk after monitor is over. My major concern is
that how big the memory should I create and how long can I log before
memory overflow?

Your help is of great help and highly appreciated!

Best regards,

-Haijin Yan