On Thu, 15 Dec 2005, Jim C. Nasby wrote:
>
> I have access to both some (SLOW) ultra5's and a machine running
> opensolaris on AMD if testing there would help. I'll need a pointer to a
> patch and test-case though...
>
Thanks! I've patched the program with the following changes:
(1) add gcc-mingw support;
(2) move the check_sort() out of do_sort() - the previous program is
actually measuring the time of qsort plus result verification. Though that
one "fairly" add equal cost to every competitor, which will not affect the
confidence of the result, it is a defeat, sorry about that;
The new results with SunOS and Windows tests are published at the same
place: http://www.cs.toronto.edu/~zhouqq/postgresql/sort/sort.html
As Luke suggested, BSD seems a good choice for scalable and stable
consideration. But I also sent an email to the author of qsortG, and he
might take a look at the small-range performance problem during the
holiday. So if he can help that, then we will have another candidate.
By the way, I do spend some time on fighting the win32 gettimeofday()
emulation. I would suggest adding a comment like "don't use this method in
windows to get high precision time, use elapsed_time() instead" ...
Regards,
Qingqing