On Wed, Mar 11, 2009 at 11:11 PM, Torsten Anders < [email protected]> wrote:
> Dear all, > I would like to do parallel search where the script contains objects. Using > objects in spaces works great in non-distributed search. The advantage is > simply an incremental definition of highly complex data structures by class > hierarchies. Yes, it is less efficient because it drains more memory. > > I just tried using my objects with parallel search and ran into > problems. Does anyone know whether parallel search allows at all for objects > in the spaces? Should these be cached, stationary or sequential asynchronous > stationary objects? > > Thank you very much! > > Best > Torsten > > PS: I am using Mozart 1.3.2 as parallel search does not work in 1.4.0. > Are those objects global? If that's the case, then they are stateless (or at least, you don't modify them from inside computation spaces). The distribution of objects in Mozart 1.3.2 is inadequate in this situation. If you really need to use objects, I suggest you find a way to serialize their state into a record or a chunk, and make sure the result does not refer to any object. The trick is then to recreate objects from the serialized stuff inside the functor that is given to the parallel search engine. By the way, do you use attributes in those objects? I ask this question because attributes and features are handled differently by the distribution: features are copied, while attributes use a dedicated protocol for state (which is not efficient if you have many concurrent reads). Would this be a possible approach for you? Cheers, raph
_________________________________________________________________________________ mozart-users mailing list [email protected] http://www.mozart-oz.org/mailman/listinfo/mozart-users
