In this lesson we shall consolidate what we have learnt so far about MPI. We shall also cover a parallel computational paradigm, which is variously referred to as ``hungry puppies'', ``bank queue'', ``master-slaves'', ``task farm'', ``job queue'', and ``self-scheduling''. My preferred name is ``bank queue'', because this is how most banks operate in Australia. But on my recent visit to Germany, for example, I have noticed that their banks don't use this practice, so the analogy may be lost on them.
This paradigm suits especially heterogeneous farms, where we may have to utilise a number of different processors. As we have seen in the previous example, distributing tasks of equal size on such a farm may result in great delays on every barrier operation, and the execution of the program will be slowed down to the speed of the slowest processor. The ``bank queue'' paradigm overcomes this difficulty.