[coyotos-dev] Accept GCJ into the tool chain?

[Jonathan S. Shapiro]

On Tue, 2006-11-28 at 14:59 -0500, Jonathan S. Shapiro wrote:

> First, those benchmarks are dominated by inner computational loops.
> Depending on the benchmark, what we are seeing is a direct measure of
> compute efficiency or memory access efficiency.

Bother. I didn't state the implication of these two statements, which
is:

  As such, these benchmarks are *conservative* predictors of relative
  performance disparities in programs that are not I/O bound.

Now, concerning I/O bound programs, we are all prone to think that if
the program is I/O bound it doesn't matter how slow it runs. This is
nonsense, because of course that longer execution time is depriving
other programs the ability to execute. A slow I/O-bound program may not
be perceptibly slower in its own right, but it has a systemic cost.

shap