I do not know how the program works and for that I apologize in advance !!
To take advantage of the power of two or more computers I think it would be useful to have the potential number generator search fields be able to be divided into segments.
so we would give each computer a separate field of research for the same data so we could find solutions to difficult programs
what is your opinion?
The algorithm can only be run on a single thread. You can open more FET programs with different random seeds and obtain different solutions at the same time.
Generally speaking, the FET algorithm works very well, and if a timetable is solvable, FET should be able to solve it. Most of the time problems occur when:
1) The timetable is impossible by itself. Sometimes the real restrictions of the timetable (for example teachers who teach in many schools) make the timetable impossible.
2) The user has not optimally expressed his dataset in terms of FET. If the problem is not optimally expressed, "invisible" constraints are introduced, and might result in an impossible timetable.
3) The user introduces not really necessary constraints for his problem. Constraints seem to be an easy part of FET, regarding their adding to the timetabling problem. But, it's just their adding procedure that is easy; not the satisfaction or the constraints themselves. Adding a new constraint in FET is as easy as changing the font in a text editing application, but the result is vastly different and might be catastrophic, since it might make the timetable impossible to be solved.
Vangelis.
Thanks for the answers ...
I made some very useful conclusions
Liviu, are there any recommendations on how to select the initial seed? For instance, if I've got a cluster with 100 nodes, I could theoretically start 100 instances of FET (or even more if I'm counting multicore CPUs). Would it be beneficial to equally distribute the seeds over a given range? Or is it only just necessary to ensure that each seed is unique?
Just ensure that each seed is unique (X or Y or both are different, even by 1).
The pseudo random number generator is taken from Knuth's TAOCP volume 2. It has a period of approximately 2^62, after which you will reach the initial random seed (improbable case). Hopefully the numbers in between will be randomly distributed. If you start the same identical file with the same random seed, you will obtain the same result.
FET uses a random number each time it needs to place randomly an activity or to choose a random number.