I can't say that I agree with your conclusion that parallel algorithms
aren't very useful for general applications nor that manually multithreading applications is the best option. On the contrary, I think it's going to be very important that the average programmer has high-level generic tools for working with parallel algorithms for even seemingly trivial applications in the not-too-far future. Coincidentally, my summer of code project for Boost, Boost.Act, is all about parallel algorithms along with active objects and simple asynchronous function calls. You can see the full proposal at http://2006.planet-soc.com/node/201 . Be sure to check out the included preliminary documentation for code samples and rationale. Right now, in terms of the algorithms side of the project, my library just allows you to easily work with a few common algorithms such as for_each, copy, generate, and fill, with the execution model able to be toggled via policies, though I plan to have all or nearly all standard algorithms implemented by the end of the summer. At this point in time I do not have any plans for parallel containers for the immediate future, but once everything else is completed, I would definately like to pursue them as a next step. Currently, my library is implemented using a combination of Boost.Threadsand OpenMP. I've read up a bit on STAPL thanks to this thread and I'm looking into the possibility of using it internally in future versions of the project, pending their willingness to release the source. -- -Matt Calabrese