2012/10/30 alexgian <alexg...@blueyonder.co.uk> > On 30 October 2012 20:22, Gustavo Gutierrez > <ggutier...@atlas.puj.edu.co>wrote: > > Just remember that the granularity that we can achieve (due to the current >> design decisions in gecode) is to de point of parallelism oin the search >> and *not* in the constraint propagation. >> >> Regards, >> Gustavo >> > > Thanks for clearing that up. > Oh well, something for Mozart 3 then ;-) > > Well I would say that in most of the cases you want your filtering algorithms to be as fast as possible and to prune as much as they can. However it is already the case in several propagators that you trade pruning for time and in the long term you still profit. Actually this is the reason in gecode for being able to specify different consistencies when you post constraints. Using parallelism inside the filtering algorithms is something that is possible in the current gecode implementation but you still pay the price of synchronization (because your propagator still needs to satisfy gecode requirements). In the case of distribution on a network of computers that synchronization is potentially more expensive. In any case it would be nice to find examples of practical problems in which all these costs are superseded by the gains in propagation.
> Good luck on the implementation integration. I have to say > "Search.parallel" or equivalent is absolutely indispensable in my opinion. > > I would also like to see this feature again. > Does this mean that "space" as a first class Mozart-Oz citizen, is > actually implemented by the gecode library rather than as a "native" > mozart-oz construct? Will it still support "choice" points as per CTM ch. > 9? > That means that conceptually there is only one space as described in CTM. Internally that space contains a gecode space that is in charge of constraint propagation. It is worth to say that gecode spaces are a specialization of "mozart spaces" that are fine tunned for constraint programming. Both of them were designed by Christian: computation spaces in "oz-light" are the ones proposed in his PhD thesis and have been available in Mozart since the beginning. Regards, Gustavo
_________________________________________________________________________________ mozart-users mailing list mozart-users@mozart-oz.org http://www.mozart-oz.org/mailman/listinfo/mozart-users
