Attached are the png files I refer to in this post. Check out the blog version to see it all in context http://www.well-typed.com/blog/60 On Thu, Oct 06, 2011 at 16:58:47 +0100, Eric Y. Kow wrote:
First, the bigger picture: why do we have another way of doing parallel programming in Haskell when we already have parallel arrays and Strategies? Part of the answer is parallelism is still a research topic, with new ideas coming out from time to time and some old ideas slowly withering away. For another part of the answer, it may help to think in terms of a trade-off between implicitness and performance, in other words, between Easy and Fast.
...
But this doesn't really tell us all that much about how to use this library. For now we come back to our word *dataflow*. The dataflow model treats program as a little black box, an input goes in, an output comes out, and what happens in between is immaterial:
The program is represented as a directed graph (a *dataflow network*), with each node representing an individual operation (black boxes in their own right), and connections between the nodes expressing dependencies. In the graph above, `g` and `h` depend on the output of `f` (the same output - there is only one) and in turn, `i` depends on the output of both `g` and `h`.
-- Eric Kow http://erickow.com
