[Ksummit-2008-discuss] DTrace

Theodore Tso tytso at mit.edu
Sat Jun 28 16:06:45 PDT 2008


On Fri, Jun 27, 2008 at 08:38:39PM -0700, Ulrich Drepper wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Theodore Tso wrote:
> > So to the extent that SystemTap can't do this thing that DTrace does
> > quite well, it really can't be marketed as a replacement or equivalent
> > of DTrace; it simply isn't right now.
> 
> That's nonsense.  It is exactly the goal.  The technology is very much
> available for this.  The problem is that not enough people with
> knowledge of the kernel has contributed their knowledge.

It is the goal, but it is not ready *now*; and that's because without
the scripts, it's not useful for system administrators.  The
assumption that the Systemtap developers seem to have was that kernel
and userspace infrastructure was the hard part, but the scripts are
easy.  Which is fine for them, I suppose, because they disclaim all
responsibility for it, and claim it's for the kernel developers to
develop those scripts.

But if that's true, why is it that Systemtap developers haven't made
it easy for kernel developers to install and use systemtap?  All of
the work by them has seemingly been optimized towards Systemtap
pre-installed in enterprise distributions.  Systemtap doesnt have
regular releases except when it is seemingly convenient for Red Hat to
their packaging convenience.  Because releases are so rarely done, and
because the wiki doesn't refer to systemtap versions (because
snapshots are so rarely done), users are invited to take development
snapshots of Systemtap's git repository.  (The download page doesn't
even have links to the "stable releases" on the ftp server.
Brilliant.)  They then discover that Systemtap depends on elf support
libraries, but the pre-installed ones on their distribution won't do,
because you need a Systemtap-specific patch, which for some unknown
reason has never been merged into the elf libraries for the better
part of two years.

But maybe as more people try using it, they'll discover some of these
rough edges, and will start trying to fix it.  Every couple of months,
I've tried using it, and because it as so many rough edges, I've
normally found it less work to debug the kernel using manual methods
rather trying to make Systemtap work on my system and with my kernel
development workflow.

							- Ted


More information about the Ksummit-2008-discuss mailing list