Re: Proposal of collaboration with SystemTap and LKST

[William Cohen <wcohen at redhat dot com>]

Masami Hiramatsu wrote:
> Hello, all.
>
> On this SystemTap ML, I noticed the hook point of LKST had been
> mentioned in January this year.
> I am very interested in this SystemTap as one of the developers of LKST.
> I would like to contribute to this project by making the best use of the
> know-how of LKST development.
> Especially, LKSTLogTools that we released recently would be very useful
> to achieve the kernel performance evaluation on the SystemTap.
> These tools support the performance analysis by processing the log that
> LKST collected. And you can download it from following URL.
> http://sourceforge.net/projects/lkst/
> My idea is to collect information on the performance by using SystemTAP
> and LKST, and to process them by LKSTLogTools later.
> I will develop TAPSets compatible with hook points about LKSTLogTools
> and LKST.

Hi,

I looked over the LKST hook points earlier in the year. LKST collects 
some useful information. We definitely would be interested in 
collaborating with LKST developers and making this type of 
instrumentation available in SystemTap.  Also insight into how 
particular analysis and instrumentation in LKST has proved useful would 
be greatly appreciated.

We are working to provide good descriptions of the various 
instrumentation sets. The URL below has a template for the description 
of the instrumetation. It would be nice to write out the LKST probes 
using this templates. Not all the fields needs to be filled out on the 
templates.

http://sources.redhat.com/ml/systemtap/2005-q2/msg00141.html

Also prototyping examples of the LKST instrumentation using the runtime 
library would be good to make make sure that the runtime library can 
support the interesting pieces of instrumentation.  Similarly, feedback 
 on how probes locations are described in SystemTap would be very 
useful. When I examined LKST there appeard to be some probe points that 
might present some difficulty in describing in a portable method.

We would like to minimize the amount of code written in raw C code and 
transfer instrumentation into the scripting language as the script 
compiler becomes functional.


> By the way, the hook points newly added with LKSTLogTools are different
> from the one that became a topic in this mailing list before. These new
> hook points provide many useful information for performance analysis.
> Please try to use LKSTLogTools. It would be greatly appreciated if some
> comments could be gotten.

Thanks for the pointer for LKSTLogTools. I just downloaded it this morning

-Will