> > 
> > Tell the cache and the tlb that, they will laugh at you for hours.
> 
> You seem to think if you keep saying this mantra over and over again
> it will mean something.
> 
The cache and tlb, and friends are very very important (specifically
conserving them.)  However, in *BSD, the kernel doesn't start taking
that much time running real workloads until the workloads get to be
very large.  That is where the algorithmic optimizations (incl cache
footprints, etc) become important.  However, the overhead of syscalls,
etc are generally so small as to be insignificant (in *BSD and Linux.)

So, it is still my position that the kernel overhead mostly becomes
significant when the system is under load...  Benchmarks like lmbench just
don't measure the system under those conditions.  That is NOT to say
that lmbench isn't useful though.  It is a valuable tool for finding
potential problems -- but ONLY potential problems.  It is also very
useful for profiling kernels with a synthetic and simple workload.

As often as not, I use lmbench to act as a sythetic workload for profiling,
and practically ignore the results of the benchmark run.  Lmbench is
indeed a wonderful tool, but is kind of like a screwdriver -- it doesn't
pound nails well, but it is excellent at what it does well :-).

John