On May 26, 5:09 pm, Jussi Piitulainen <jpiit...@ling.helsinki.fi> wrote: > > A light-weighter way is to have each task end by assigning the next > task and returning, instead of calling the next task directly. When a > task returns, a driver loop will call the assigned task, which again > does a bounded amount of work, assigns the next task, and returns. > Tasks can even pass parameters in the same way. > Yes, that's true - there are a number of ways of making it linear.
What I'm particularly pleased about with my method is the parallelism that it achieves - with so little effort! The simulation is going to be computationally intense and this is going to make sure that the CPUs are all giving it their best shot. When I run this on my macbook, the python interpreter takes over 140% of CPU - with a bit of fine- tuning, it should be possible to have more concurrent threads and to use the four cores optimally. Naturally I'll need to be careful with the concurrency, but this is so simple and clean that it should be easy to avoid the main problems with accessing the same variables. -- http://mail.python.org/mailman/listinfo/python-list
