On Mon, 12 Feb 2018 17:35:13 -0500 Ken Hornstein <k...@pobox.com> wrote:
Ken Hornstein writes:
> >As I showed, the perf loss via
> >using GC with nmh is tolerable and this is without doing any
> >cleanup or code improvement based on profiling.
> Errr ... were there some other messages where this assertion was made?
> Because my takeway was that with garbage collection it was MUCH MUCH
> worse. Like, "Oh, hell no, of course we're not doing GC!" worse.
There was a fair bit of variability but it was no worse than
twice as slow and often close to par. I will try to check this
out on a mac and see if I get more consistent results. But to
me even twice as slow as a starting point would be fine. The
most time consuming op is pick and its performance is more
dependent on file reading than anything else.