On Hierarchy
A nice line:
Hierarchies. Cannot think within them. Cannot think without them.
Arthur Koestler in "The Ghost in the Machine" (amazon.com, amazon.co.uk) talks about arborisation and reticulation as complementary. Hierarchical structuring is powerful, but a closed hierarchy is restrictive. The leaves of the hierarchy (at least) must connect with those of other hierarchies to form powerful structures.
I think Koestler's insight provides quite a conceptual leg-up when thinking about the composition of system representations ("models") with which to drive simulations. Considered as gross components, subsystem representations are assembled in a composition hierarchy which is a straightforward tree. The leaves of this tree are atomic components which are defined in some way other than by composition. Composites do something useful because they define connections between the inputs and outputs of their component parts, and ultimately this alsways means connecting the inputs and outputs of the atomic leaf components. Thus the modeller is simultaneously defining a tree and a network.
When a single-purpose model is being constructed, this will result in a single tree. We can imagine, however, that long-running systems expose the data flows at their leaves in some way (publish and subscribe). In this case it should be possible to define new systems which connect to these flows, and then we start to define a forest of individual trees.
Comments