On Wed, 17 Nov 2004 11:24:46 +0100, Erich Focht wrote:
> On Wednesday 17 November 2004 10:49, Guillaume Thouvenin wrote:
> > Selon "Philip J. Mucci" <mucci@...>:
> > > Why does ELSA use a kernel module to do something that could easily be
> > > done in user space by trapping fork/exec in a LD_PRELOADed library
> > > with a constructor?
> > - What is the performance overhead compared to a module (context
> > switch..) ?
> > - Are we sure to get all forks?
> With LD_PRELOAD you won't catch statically linked binaries. This
> provides users a way to run unaccounted processes and would make
> PAGG useless. The kernel patch is necessary AFAICS.
Yeah - LD_PRELOAD is an interesting hack but not a solution to
standard system accounting. A kernel patch or module is the
correct solution here.