dtrace discuss - Sep 2006 - Comments on a Systemtap vs Dtrace comparison.

Not sure if this document will be shown to any dtrace users but here
is a link if you want to comment on it

http://sources.redhat.com/systemtap/wiki/SystemtapDtraceComparison

I have went line by line and commented on its content and in
accuracies and posted it both to the systemtap email  list and my
blog, you can see the blogged version at

http://uadmin.blogspot.com/2006/09/going-line-by-line.html

but feel free to post your comments to the systemtap mailing list or
as comments to my blog.

the Systemtap mailing list is at "SystemTAP"
<systemtap at sources.redhat.com> not sure if you need to subscribe to
post or not.



James Dickens
uadmin.blogspot.com

I''m not familiar enough with either (let alone both) to comment on the
accuracy.  But less hyperbole and (sorry) better grammar and spelling,
however irrelevant to the technical content, would nevertheless
encourage a more receptive audience.

It is my impression that the existance and utility of Dtrace must have
encouraged the development of Systemtap.  Further, an example (even
if only to be used for ideas rather than copy/paste) should have let
Systemtap not only develop faster (as it may have - not sure what Dtrace
looked like at one year old), but should by rights have let it aim significantly
higher.  It''s not clear to me that it did that.

Processor ports are interesting, but to my mind, above two processor
ports, OS ports are more interesting, in terms of what they show about
the adaptability of the design.

Going seamlessly across kernel, userland, and even those interpreted
environments for which probes have been created, esp. on existing
binaries not specially compiled, strikes me as indispensible.  I gather
that only one of the two can do that right now.

In any case, competition is good.  Hopefully, Systemtap will come up
with some good ideas that Dtrace can implement (if differently).  It
appears nearly certain that the reverse has already happened quite a bit.
 
 
This message posted from opensolaris.org

On 9/12/06, Richard L. Hamilton <rlhamil at smart.net>
wrote:
> I''m not familiar enough with either (let alone both) to comment on
the
> accuracy.  But less hyperbole and (sorry) better grammar and spelling,
> however irrelevant to the technical content, would nevertheless
> encourage a more receptive audience.
>
> It is my impression that the existance and utility of Dtrace must have
> encouraged the development of Systemtap.  Further, an example (even
> if only to be used for ideas rather than copy/paste) should have let
> Systemtap not only develop faster (as it may have - not sure what Dtrace
> looked like at one year old), but should by rights have let it aim
significantly
> higher.  It''s not clear to me that it did that.
>
I have been an active participant in there mailing list, and have
commented on ther mistakes as well as filing numerous bug reports to
their project. I have also invited them to join this mailing list so
that they could discuss the problems they are having, with people that
designed and work with the same type of technology for years now.

All offers have been refused. And apparently from the chart and its
content most of them have little knowledge of what dtrace can do and
how safe it is.

Usually everytime I file a bug report, its first debated that it
really happened, and then they say  that its most likely a user error,
then finally after they get through all that debate, then they take a
serious look at the bug and decide that it actually exists and is a
bug. This is not a unique experience from the Linux community. It
happens with just about every bug ever submitted by a user.

I really don''t know why I keep trying to help them out. But I do and
continue to do so.

James Dickens
uadmin.blogspot.com

> Processor ports are interesting, but to my mind, above two processor
> ports, OS ports are more interesting, in terms of what they show about
> the adaptability of the design.
>
> Going seamlessly across kernel, userland, and even those interpreted
> environments for which probes have been created, esp. on existing
> binaries not specially compiled, strikes me as indispensible.  I gather
> that only one of the two can do that right now.
>
> In any case, competition is good.  Hopefully, Systemtap will come up
> with some good ideas that Dtrace can implement (if differently).  It
> appears nearly certain that the reverse has already happened quite a bit.
>
>
> This message posted from opensolaris.org
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org
>

> I have been an active participant in there mailing
> list, and have
> commented on ther mistakes as well as filing numerous
> bug reports to
> their project. I have also invited them to join this
> mailing list so
> that they could discuss the problems they are having,
> with people that
> designed and work with the same type of technology
> for years now.
> 
> All offers have been refused. And apparently from the
> chart and its
> content most of them have little knowledge of what
> dtrace can do and
> how safe it is.
>
> Usually everytime I file a bug report, its first
> debated that it
> really happened, and then they say  that its most
> likely a user error,
> then finally after they get through all that debate,
> then they take a
> serious look at the bug and decide that it actually
> exists and is a
> bug. This is not a unique experience from the Linux
> community. It
> happens with just about every bug ever submitted by a
> user.
I''d say that both their inaccurate (at least I suspect it is too,
although more based on the questionable match of their
characterization of Dtrace with what I''ve read about it elsewhere
than on anything else) comparison, and their attitude about
bugs have something in common: arrogance.

Most users _are_ stupid (or let''s be nice: ignorant), and most
bug reports _are_ bogus.  But belaboring that point also
discourages _real_ bug reports, which represent opportunities
for improvement.  It sounds like they have yet to learn that...
> I really don''t know why I keep trying to help them
> out. But I do and
> continue to do so.
Firm but tactful...and either they''ll learn, or they won''t.
You may be responsible for giving them something to
listen to, and for _not_ giving them trivial excuses to
turn away...but if you do those things, you''re not responsible
for whether or not they listen.
 
 
This message posted from opensolaris.org

James, your comparison is wonderful.

IMHO the number one thing that stands out at a 30,000 foot comparison of both at
this point is the statement that''s clearly made directly on the
SystemTap project page:

  [b]SystemTap is still under rapid development, so it is not appropriate to use
it on production systems.[/b]

So despite an assertion in the recent News.com article that "SystemTap is
similar to Sun Microsystems'' DTrace, which enables administrators to
probe systems, as they run, to scrutinize software for bottlenecks,"
obviously with the statement above the comparison comes to a screeching halt.
 
 
This message posted from opensolaris.org

On Tue, Sep 12, 2006 at 10:35:15PM -0500, James Dickens
wrote:
> Not sure if this document will be shown to any dtrace users but here
> is a link if you want to comment on it
> 
> http://sources.redhat.com/systemtap/wiki/SystemtapDtraceComparison
Wow, is that comparison wrong.  On so many points.

When two independent non-Sun teams have ported DTrace you can''t say
that
DTrace''s community is "Sun internal" -- it''s _open_,
period.

And that''s just a start.  On technical details they got it wrong in
many
places, as your blog notes.
> I have went line by line and commented on its content and in
> accuracies and posted it both to the systemtap email  list and my
> blog, you can see the blogged version at
> 
> http://uadmin.blogspot.com/2006/09/going-line-by-line.html
> 
> but feel free to post your comments to the systemtap mailing list or
> as comments to my blog.
I don''t know what they meant by binary tracing either, BTW, but if they
meant tracing random instructions, instead of function entry/return,
well, DTrace does that in userland, and if they meant tracing binary
data, well, DTrace does that too.  What else could they have meant?

BTW, DTrace does not handle traps when copying in userland data from
kernel probe points, which is why one typically has to grab, say, the
path argument to open(2) in the return probe (since by then the path had
to have been copied it by the system call any page faults must have been
handled).  A fine point, but worth noting.

Nico
--

The online comparison obviously was slanted to make Systemtap sound superior to
DTrace. I think that Sun should "officially" do the comparison on
it''s website.

As an instructor that teaches the new features of Solaris 10 class for SunEd,
when I first read the online comparison my initial feelings were that Linux now
has a better tool than Solaris.

But by reading James (informing) comparison I can see that it is nothing more
than mis-information. So let''s spread the word.DTrace stile rules!

:)


Merle Ilgenfritz
Ilgenfritz Consulting, LLC
Crossville, TN 38555



>  -------Original Message-------
>  From: William J. Kucharski <William.Kucharski at Sun.COM>
>  Subject: [dtrace-discuss] Re: Comments on a Systemtap vs Dtrace
comparison.
>  Sent: Sep 13 ''06 08:34
>  
>  James, your comparison is wonderful.
>  
>  IMHO the number one thing that stands out at a 30,000 foot comparison of
both at this point is the statement that''s clearly made directly on the
SystemTap project page:
>  
>  ??[b]SystemTap is still under rapid development, so it is not appropriate
to use it on production systems.[/b]
>  
>  So despite an assertion in the recent News.com article that
"SystemTap is similar to Sun Microsystems'' DTrace, which enables
administrators to probe systems, as they run, to scrutinize software for
bottlenecks," obviously with the statement above the comparison comes to a
screeching halt.
>  
>  
>  This message posted from opensolaris.org
>  _______________________________________________
>  dtrace-discuss mailing list
>  dtrace-discuss at opensolaris.org
>

> > Not sure if this document will be shown to any dtrace users but here
 > > is a link if you want to comment on it
 > > 
 > > http://sources.redhat.com/systemtap/wiki/SystemtapDtraceComparison
 > 
 > Wow, is that comparison wrong.  On so many points.

I also love the use of "not yet" for "no" in the SystemTap
columns.

--
meem

So the compelling statistics, which are, unfortunately, impossible to 
track...

Number of bugs found using the tool, IN PRODUCTION:
DTrace - hundreds or thousands
SystemTap - almost zero, possibly zero

Number of positive workload performance improvement experiences, IN 
PRODUCTION:
DTrace - Thousands
SystemTap - almost zero, possibly zero

Number of performance or behavior escalation issues resolved, IN PRODUCTION:
DTrace - hundreds, perhaps thousands
SystemTap - almost zero, possible zero

Number of system crashes/hangs/panics induced IN PRODUCTION:
DTrace - almost zero, may well be zero.
SystemTap - who knows...we''ll see when it''s in production...

To heck with the feature comparison matrix. These 4 statistics speak 
volumes....

/jim



                                     

Merle Ilgenfritz wrote:
> The online comparison obviously was slanted to make Systemtap sound
superior to DTrace. I think that Sun should "officially" do the
comparison on it''s website.
>
> As an instructor that teaches the new features of Solaris 10 class for
SunEd, when I first read the online comparison my initial feelings were that
Linux now has a better tool than Solaris.
>
> But by reading James (informing) comparison I can see that it is nothing
more than mis-information. So let''s spread the word.DTrace stile rules!
>
> :)
>
>
> Merle Ilgenfritz
> Ilgenfritz Consulting, LLC
> Crossville, TN 38555
>
>
>
>
>   
>>  -------Original Message-------
>>  From: William J. Kucharski <William.Kucharski at Sun.COM>
>>  Subject: [dtrace-discuss] Re: Comments on a Systemtap vs Dtrace
comparison.
>>  Sent: Sep 13 ''06 08:34
>>  
>>  James, your comparison is wonderful.
>>  
>>  IMHO the number one thing that stands out at a 30,000 foot comparison
of both at this point is the statement that''s clearly made directly on
the SystemTap project page:
>>  
>>    [b]SystemTap is still under rapid development, so it is not
appropriate to use it on production systems.[/b]
>>  
>>  So despite an assertion in the recent News.com article that
"SystemTap is similar to Sun Microsystems'' DTrace, which enables
administrators to probe systems, as they run, to scrutinize software for
bottlenecks," obviously with the statement above the comparison comes to a
screeching halt.
>>  
>>  
>>  This message posted from opensolaris.org
>>  _______________________________________________
>>  dtrace-discuss mailing list
>>  dtrace-discuss at opensolaris.org
>>  
>>     
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org
>

And that my friends says it all!

[Put that as an addendum in the blog!]

:)


Merle Ilgenfritz
Ilgenfritz Consulting, LLC
Crossville, TN 38555



>  -------Original Message-------
>  From: Jim Mauro <James.Mauro at Sun.COM>
>  Subject: Re: [dtrace-discuss] Re: Comments on a Systemtap vs Dtrace
comparison.
>  Sent: Sep 13 ''06 13:01
>  
>  So the compelling statistics, which are, unfortunately, impossible to
>  track...
>  
>  Number of bugs found using the tool, IN PRODUCTION:
>  DTrace - hundreds or thousands
>  SystemTap - almost zero, possibly zero
>  
>  Number of positive workload performance improvement experiences, IN
>  PRODUCTION:
>  DTrace - Thousands
>  SystemTap - almost zero, possibly zero
>  
>  Number of performance or behavior escalation issues resolved, IN
PRODUCTION:
>  DTrace - hundreds, perhaps thousands
>  SystemTap - almost zero, possible zero
>  
>  Number of system crashes/hangs/panics induced IN PRODUCTION:
>  DTrace - almost zero, may well be zero.
>  SystemTap - who knows...we''ll see when it''s in
production...
>  
>  To heck with the feature comparison matrix. These 4 statistics speak
>  volumes....
>  
>  /jim
>  
>  
>  
>  ????????????????????????????????????
>  
>  Merle Ilgenfritz wrote:
>  > The online comparison obviously was slanted to make Systemtap sound
superior to DTrace. I think that Sun should "officially" do the
comparison on it''s website.
>  >
>  > As an instructor that teaches the new features of Solaris 10 class
for SunEd, when I first read the online comparison my initial feelings were that
Linux now has a better tool than Solaris.
>  >
>  > But by reading James (informing) comparison I can see that it is
nothing more than mis-information. So let''s spread the word.DTrace
stile rules!
>  >
>  > :)
>  >
>  >
>  > Merle Ilgenfritz
>  > Ilgenfritz Consulting, LLC
>  > Crossville, TN 38555
>  >
>  >
>  >
>  >
>  >??
>  >>??-------Original Message-------
>  >>??From: William J. Kucharski <William.Kucharski at Sun.COM>
>  >>??Subject: [dtrace-discuss] Re: Comments on a Systemtap vs Dtrace
comparison.
>  >>??Sent: Sep 13 ''06 08:34
>  >>??
>  >>??James, your comparison is wonderful.
>  >>??
>  >>??IMHO the number one thing that stands out at a 30,000 foot
comparison of both at this point is the statement that''s clearly made
directly on the SystemTap project page:
>  >>??
>  >>????[b]SystemTap is still under rapid development, so it is not
appropriate to use it on production systems.[/b]
>  >>??
>  >>??So despite an assertion in the recent News.com article that
"SystemTap is similar to Sun Microsystems'' DTrace, which enables
administrators to probe systems, as they run, to scrutinize software for
bottlenecks," obviously with the statement above the comparison comes to a
screeching halt.
>  >>??
>  >>??
>  >>??This message posted from opensolaris.org
>  >>??_______________________________________________
>  >>??dtrace-discuss mailing list
>  >>??dtrace-discuss at opensolaris.org
>  >>??
>  >>????
>  > _______________________________________________
>  > dtrace-discuss mailing list
>  > dtrace-discuss at opensolaris.org
>  >??
>

On Wed, Sep 13, 2006 at 02:11:31PM -0400, peter.memishian at sun.com
wrote:
> 
>  > > Not sure if this document will be shown to any dtrace users but
here
>  > > is a link if you want to comment on it
>  > > 
>  > >
http://sources.redhat.com/systemtap/wiki/SystemtapDtraceComparison
>  > 
>  > Wow, is that comparison wrong.  On so many points.
> 
> I also love the use of "not yet" for "no" in the
SystemTap columns.
Consider if you will, my comparison of DTrace to pork:

			DTrace		pork
------------------------------------------------
kosher			yes		no
always free of disease	yes		no
open source		yes		maybe -- license terms unclear
rate of evolution	high		glacial
delicious		not yet		yes

I''m not sure we should take this comparison any more seriously than the
pork council takes mine.

Adam

-- 
Adam Leventhal, Solaris Kernel Development       http://blogs.sun.com/ahl