SystemTap Dissertation

SystemTap Dissertation

[swb6]

Hi,
Im an Undergraduate computer science student and have been very interested
in the emergence of new kernel tracing technologies and I have based my
dissertation on systemTap.

The application i have designed allows a user to deploy system tap scripts
to a number of hosts across a network, and retrieve and store results from
each system, with scope for real time graphing of performance (so far only
file system read writes).

I would like to offer the user the ability to create there own scripts by
offering a list of possible probe points that they can access.

I have see that something similar is possible in D trace and can be be
achieved via the command 'dtrace -l -P <provider>
which will list all the probes available in that providers area e.g.
syscall provider.

Is anything like this available with systemTap, If I could provide a list
of probe points for a user specified area of the kernel I could reduce the
learning curve to anyone new to systemTap and kernel tuning/debugging
(this includes myself)

Sorry to hijack this list with such a basic question

Steve

Re: SystemTap Dissertation

[Josh Stone]

On 03/15/2010 02:20 PM, swb6@aber.ac.uk wrote:
> Hi,
> Im an Undergraduate computer science student and have been very interested
> in the emergence of new kernel tracing technologies and I have based my
> dissertation on systemTap.
> 
> The application i have designed allows a user to deploy system tap scripts
> to a number of hosts across a network, and retrieve and store results from
> each system, with scope for real time graphing of performance (so far only
> file system read writes).

Interesting.  I think there are a few other projects around real-time
graphing, including the stap grapher we have in the source tree.  It may
be helpful for you to compare ideas.

And please fell free to share your progress and the result with us!

> I would like to offer the user the ability to create there own scripts by
> offering a list of possible probe points that they can access.
> 
> I have see that something similar is possible in D trace and can be be
> achieved via the command 'dtrace -l -P <provider>
> which will list all the probes available in that providers area e.g.
> syscall provider.
> 
> Is anything like this available with systemTap, If I could provide a list
> of probe points for a user specified area of the kernel I could reduce the
> learning curve to anyone new to systemTap and kernel tuning/debugging
> (this includes myself)

Sure - you can use wildcards with the -l option to get lists, e.g.:

  stap -l 'syscall.*'
  stap -l 'kernel.function("sched*")'
  etc.

There is also -L which provides the same listing along with the
variables available at each point.

> Sorry to hijack this list with such a basic question

No problem, this is a perfectly valid question for the list.


Josh

Re: SystemTap Dissertation

[Aaron Luchko]

On Mon, 2010-03-15 at 21:20 +0000, swb6@aber.ac.uk wrote:
> Hi,
> Im an Undergraduate computer science student and have been very interested
> in the emergence of new kernel tracing technologies and I have based my
> dissertation on systemTap.

Cool, I'm a grad student doing a course project trying to build a tool
around systemtap.

It's still very preliminary but my objective is to record and save a
series of realtime metrics on a particular host.

It will then display the metrics to the user through gnuplot. The idea
is to show "interesting" metrics as defined by any metric currently
outside it's 95% confidence interval.

The current state of the project is here

git@github.com:aluchko/SystemTap-Health-Monitor.git

It currently polls CPU usage every second and drops the results in an
sqlite database. The front end than graphs the results of whichever
process happens to have the id of 5 in the db (still very hardcoded!).

> The application i have designed allows a user to deploy system tap scripts
> to a number of hosts across a network, and retrieve and store results from
> each system, with scope for real time graphing of performance (so far only
> file system read writes).

Not sure if there's any opportunity for collaboration but you're free to
poke around for anything useful. I have a very simple systemtap
script/backend communication protocol described in the README and I do a
running calculation of variance in Metric.cxx.

-aaron