A simple search on the zilla dataset of Chase et al. (1993) using traditional heuristic settings (60 reps) performed first in serial, then in parallel 20 reps x 3 slaves.

100 seconds for search 1, down to just 48 seconds for search 2 (in parallel). YMMV

Neither of these searches found the shortest length trees btw!

Commands:
mxram 100; /* increase memory */
p zilla.tnt; /* read in the data */
hold 20000; /* increase the maximum number of trees held */
mult 60; /* perform a traditional search with 60 replications */
le; /* tree lengths */

/* parallel tnt job, called 'ajob' using 3 slaves performing 'mult 20' on each slave */
ptnt begin ajob 3 = mult 20; return ; ptnt wait . ;

ptnt get ajob; /* get data back from slaves to master */

свернуть