dtrace discuss - Nov 2005 - Some challenging questions from the Beijing OpenSolaris User Group

Hi folks,

We have been asked some challenging question on dtrace from the Beijing
OpenSolaris User Group
<http://www.opensolaris.org/os/community/os_user_groups/bjosug> meeting:

1. The dtrace action printf() will use the tty driver. If there are something
broken in that driver, does it mean that all printings are rubbish?

2. what''s the test methodology of dtrace? ie. how to prove dtrace
result is right?

I''m not sure about these two questions, can you help to find the
answer?

For more details please check out my blog <http://blogs.sun.com/JoeyGuo>. 

Any comments are appreciated!

Thanks & Best regards,

-- 
Joey Guo, Solaris x86 Engineering, http://blogs.sun.com/JoeyGuo 
Phone number: (8610) 8261-8200 (x.82245)

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/dtrace-discuss/attachments/20051102/bc651274/attachment.html>

Hi Joey,

Here are the answers to the DTrace-related questions on your blog:

   1. The dtrace action print() will use the tty driver. If there are
   something broken in that driver, does it mean that all printings are
   rubbish?

All output is done once in user-land. The actual trace() or printf()
actions record data to the DTrace buffer which is captured by the
user-land consumer (usually dtrace(1M)) every so often. So no component of
DTrace uses the tty driver directly, rather the consumer uses standard
write(2) system calls out output data.

   2. What''s the purpose of print out all the callstack in the Solaris 

The stack() action and ustack() action can be used to display the kernel
or user-land stack traces respectively. Those actions can be useful when
trying to understand where a certain activity is taking place.

   3. Performance impact of dtrace

The important aspect of this question is that the performance impact is zero
when DTrace is not in use. When DTrace is being used, the performance impact
depends largely on what probes you''re enabling, what the traced system
is
doing and what actions you take in those probes.

   4. What changes need to be made in order to port DTrace to FreeBSD/Linux?

Devon O''Dell is working on porting DTrace to FreeBSD. You can find
details
here: http://www.sitetronics.com/wordpress/

   5. what''s the test methodology of dtrace. (how to prove dtrace
   result is right?) I''m not sure about this one, can you help to find
   the answer?

We''ve developed an extensive battery of tests for DTrace that exercise
the mechanics and correctness of the DTrace framework. We hope to open
source that test suite at some point.

Adam

On Wed, Nov 02, 2005 at 10:09:39PM +0800, Joey Guo
wrote:
> Hi folks,
> 
> We have been asked some challenging question on dtrace from the Beijing
OpenSolaris User Group
<http://www.opensolaris.org/os/community/os_user_groups/bjosug> meeting:
> 
> 1. The dtrace action printf() will use the tty driver. If there are
something broken in that driver, does it mean that all printings are rubbish?
> 
> 2. what''s the test methodology of dtrace? ie. how to prove dtrace
result is right?
> 
> I''m not sure about these two questions, can you help to find the
answer?
> 
> For more details please check out my blog
<http://blogs.sun.com/JoeyGuo>.
> 
> Any comments are appreciated!
> 
> Thanks & Best regards,
> 
> -- 
> Joey Guo, Solaris x86 Engineering, http://blogs.sun.com/JoeyGuo 
> Phone number: (8610) 8261-8200 (x.82245)
> 
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org
-- 
Adam Leventhal, Solaris Kernel Development       http://blogs.sun.com/ahl