Why Dataflow machines?
Parallelism is inherent since we define all nodes with ready data as being able to be executed
- Problem is identifying which nodes have ready data and getting data to those nodes
Synchronization between multiple CPUs is done at a low level and is automatic
- Data dependencies control sequencing of instructions
Fine grain data flow machines are difficult
Have tried Hybrid data flow machines in which data flow concepts applied at process level, normal instruction flow within a process