On Fri, Nov 02, 2001 at 07:20:36AM -0500, Hubertus Franke wrote:
>
> One more. Throughout our MQ evaluation, it was also true that
> the overall performance particularly for large thread counts was
> very sensitive to the goodness function, that why a na_goodness_local
> was introduced.
>
Correct, we did notice measurable differences in performance just
from the additional (unnecessary) checks in goodness. Unfortunately,
the current version of MQ has 3 different (but similar) variations
of the goodness function. This is UGLY, and I intend to clean this
up (without impacting performance of course :).
--
Mike