I made an initial performance measurement of several parallel random-DFS
scans. I ran 5 scans with the seeds 1, 2, 3, 4 and 5. I ran them in
parallel up to a limit of 50,000 states on the Microsoft 32,000.
The results are:
* It took a total of 145 minutes to solve the boards on a Pentium III 667
MHz with Compact States. That makes it 220 board per minute.
* 16 boards were reported as intractable.
* When I ran a regular test_multi run in parallel, I noticed that it
progressed much slower than the parallel random-DFS one.
* From what I noticed the intractable boards prove hard even to different
seeds than the ones I used. However, usually one seed can be found that is
OK.
So, all in all, random-DFS proved to be a very promising methodology.
Regards,
Shlomi Fish
----------------------------------------------------------------------
Shlomi Fish shlomif_at_vipe.technion.ac.il
Home Page:
http://t2.technion.ac.il/~shlomif/
Home E-mail: shlomif_at_techie.com
If:
1. A is A
2. A is not not-A
does it imply that
1. B is B
2. B is not not-B
Received on Thu Nov 29 2001 - 02:11:03 IST