[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Status of systemtap in Debian

On 16/02/10 at 22:17 +0100, Kurt Roeckx wrote:
> On Tue, Feb 16, 2010 at 09:13:06PM +0100, Lucas Nussbaum wrote:
> > 
> > - disk space on buildds: at least 2 GiB are required to build a kernel
> >   with debuginfo. (that doesn't sound too hard to satisfy)
> There currently are packages that require more diskspace than
> that, for instance the linux-2.6 package on i386 uses 4.7 GB,
> while it uses 2.6 GB on amd64.
> I guess it requires an additional 450 MB per kernel flavor
> or maybe a multiple of that for temporary files that also
> contain that?
> The biggest package I could find in a short time seem to be:
> qt4-x11:                7818730k (8058048k latest)
> openoffice.org:         7652264k (7629368k latest)
> openjdk-6:              7152062k (7443004k latest)
> linux-2.6:              4707725k (6319992k latest)
> opencascade:            4619392k (4619392k latest)
> wxwidgets2.8:           4215131k (4297664k latest)
> ghc6:                   4015377k (4689676k latest)
> slicer:                 3680980k (3680980k latest)
> gcc-snapshot:           3663284k (3663284k latest)
> pyside:                 3548760k (3548760k latest)
> boost1.38:              3497562k (3630040k latest)
> (This is acros several arches.)
> Do you know if there is any cost in cpu time for this?

At build time, of course, generating the debug info takes more time.

At run time, I don't think that a kernel built with CONFIG_KPROBES and
CONFIG_DEBUGINFO has any overhead, at least on i386/amd64 (remember that
all other distros ship KPROBES-enabled kernels). Since it depends on the
implementation of kprobes, which is architecture-specific, it is
possible that there is an overhead on some architectures.

- Lucas

Reply to: