I am working for a real problem. It is very difficult algorithmic problem. After reducing complexity and optimizing all possible bottlenecks the last step is parallelization. There are many dependent chunks. It looks more like DAG of jobs with low branching. Each chunk can be time consuming and because it often happens that I can distinguish just few (often two) parallel local jobs and than move forward to next part, I would loose too much time waiting in Finish(). I had to edit it. I am using Pipe also, it is really helpful. Generally U++ for algorithmic problems is just great.