Graph or Task Model Of Parallel Computation
1. All parallel computation structures can be represented as dynamic hierarchical graphs where nodes are the sites of execution of units of computation, the arcs specify the dependence relations and nodes and arcs are given appropriate types and semantics. A node execution has the atomicity property.
2. Nodes may either contain a unit of computation or a graph.
3. All executions of parallel programs are traversals of instances of such graphs by appropriately defined units of work.
4. The types and semantics for nodes and arcs can be declaratively specified.
5. We will speak of static graphs which are programs and dynamic graphs which are executions of a program.
6. A parameterized static graph and an initialization determines one instance of a dynamic task graph.