[dtrace-discuss] DTrace under Linux

[Francois Dion]

On Wed, 2007-02-14 at 20:08 +0800, 陶捷 (Euler Tao) wrote:
> Is DTrace talked about in the book, "Solaris Internals. Edition 2"?
> I haven't got this book. But someone told me that this book maybe the
> most detail book about Solaris kernel code.

You are talking about:
http://www.sun.com/books/catalog/solaris_internals.xml

This is definitely the book to learn about the Solaris kernel and how
processes work in Solaris. Part two in particular:
PART TWO: The Process Model 
Chapter 2: The Solaris Process Model
Chapter 3: Scheduling Classes and the Dispatcher 
Chapter 4: Interprocess Communication 
Chapter 5: Process Rights Management



For something that is pretty much all about DTrace, take a look also at:
http://www.sun.com/books/catalog/solaris_perf_tools.xml

Solaris Performance and Tools: Dtrace and MDB techniques for Solaris 10
and OpenSolaris.

This book is *great*.  It is the companion to Solaris Internals edition
2. I've learned a good bit with this book, highly recommended.


As a side note, Jim Mauro, one of the author has just done 3 days up in
Canada:
http://www.exitcertified.com/newsandevents/eventdetails/2007/solaris_roadshow/

covering process tools and dtrace and mdb. Apparently this was quite an
impressive session, some linux users in academia and the business world
were impressed enough to want to take a look at Solaris, just because of
dtrace.

I believe this was the full length version of the presentation that
George Wilson did at Sun Tech Days Atlanta, a real nice presentation
too. I dont have a link for it, but I believe this presentation is
available in open/star office presentation format. Or maybe just in pdf.
Do take a look at that.

Francois

> 2007/2/14, Sean McGrath - Sun Microsystems Ireland
> <sean.mcgrath at sun.com>:
>         ???? (Euler Tao) stated:
>         < Thanks, I understand now.
>         <
>         < er... I've planed to read the source code of DTrace. I think
>         it may be
>         < consisted of two parts, the kernel of Dtrace and the 30K+
>         providers. 
>         < And is the main code of DTrace in a separate directory in
>         the whole kernel
>         < code tarball, e.g. on-src-b56.tar.bz2.
>         < I've searched the key word, "dtrace", in the web page,
>         < http://cvs.opensolaris.org/source/.
>         < And I was confused by the results. They looked a little
>         disordered.
>         
>         Dtrace may look 'disordered'.  You're probably finding all the
>         dtrace probes points etc that are scattered all over the place
>         in various 
>         modules, libraries etc.  The bits that really make up  the
>         dtrace framework
>         are better explained from Bryan's blog, below.
>         
>         <
>         < I have no experience on reading any Solaris's Codes yet.
>         < Can someone do me a favor? Give me some hints. 
>         
>           Dtrace was the original bit of code that was opened up
>         before the full
>           opensolaris, way back.  Bryan's blog gives us a tour of the
>         dtrace code
>           from this initial code:
>         
>              http://blogs.sun.com/bmc/date/20050125
>         
>           Its still good to go through this once in a while.
>         
>           That along with the dtrace usenix paper gives a good
>         overview of how
>           dtrace works along with where it lives in opensolaris. 
>         
>               http://www.sun.com/bigadmin/content/dtrace/dtrace_usenix.pdf
>         
>           That should help you get started :)
>         
>         Sean.
>         .
>         <
>         < Thanks!
>         <
>         < Regards
>         <
>         < T.J.
>         <
>         <
>         < 2007/2/14, Casper.Dik at sun.com <Casper.Dik at sun.com>:
>         < >
>         < >
>         < >>Why do you say that SystemTap only is done on i386?
>         < >>Does it means that, compared with DTrace, SystemTap is
>         focused more on
>         < >>hardware infomation?
>         < > 
>         < >Dynamic kernel probing tools like DTrace need to understand
>         and modify
>         < >the instruction streams in the kernel.
>         < >
>         < >>Does DTrace has two different kernel parts?
>         < >>I means for the x86 and SPARC architecture, there're
>         something different 
>         < >>inside the 2 DTraces?
>         < >
>         < >Yes; DTrace is atypical in that it requires quite intimate
>         knowledge
>         < >of how instruction streams work.  But from a user's
>         perspective this
>         < >is all hidden and you can use the same scripts on both
>         architecture.
>         < >
>         < >The DTrace architecture limits the knowledge of the
>         processor architecture
>         < >to the DTrace to the providers which need that knowledge
>         (such as the 
>         < >fbt and fasttrap provider)
>         < >
>         < >Sun has a requirement for "completeness", i.e., support for
>         all
>         < >architectures,
>         < >when putting a project in Solaris.  That's why we have
>         support for DTrace 
>         < >on all supported platforms.
>         < >
>         < >In Linux, it is possible that systemtap becomes mainstream
>         without support
>         < >for other architectures.  (Systemtap is also much more
>         limited and does 
>         < >not
>         < >make the safety guarantees DTrace makes)
>         < >
>         < >Casper
>         < >
>         
>         < _______________________________________________
>         < dtrace-discuss mailing list
>         < dtrace-discuss at opensolaris.org
>         
>         
>         --
>         Sean.
>         .
> 
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org