[Ksummit-2008-discuss] DTrace

Christoph Hellwig hch at lst.de
Fri Jun 27 09:29:14 PDT 2008


On Fri, Jun 27, 2008 at 10:23:33AM -0500, James Bottomley wrote:
> DTrace is more a piece of sun marketing coolaid which they use to beat
> us up at every opportunity.
> 
> We actually have a reasonably functional equivalent piece of technology
> called systemtap.

Actually we don't, and while there's a large amount of coolaid around
dtrace there are also various very useful bits around it.

Without going on the implementation level the obvious benefit of dtrace
is that it just works.  It comes with solaris, there is an enormous
amount of existing static trace points in the kernel, and writing
additional traces is rather easy.  Also what's actually most important
for most dtrace users I've seen is the ability to trace through
userspace programs, and not actually kernel code which is something
systemtap currently doesn't do at all, and something which to work
should not require an out of tree metacompiler to generate kernel
modules.

An additional advantage for dtrace is that they actually have modified
lots and lots of userspace to support their userspace tracepoints, like
the postgres example Matthew mentioned or the X server or Java.
Duplicating all the this would be outright stupid, so we'd better have
an API-level compatible implementation, even if our underlying
implementation is different.


More information about the Ksummit-2008-discuss mailing list