Abstract
In the controlflow paradigm, based on the finite automata theory, one writes a program in order to control the flow of data through the hardware. In the dataflow paradigm, one writes a program in order to configure the hardware. Then, the question is what moves data through the hardware, if that is not a stored program? Ideally, in the dataflow paradigm, data get moved by the voltage difference between the system input and the system output, with flip-flops and register barriers in-between (FPGAs), or without these barriers (referred as ultimate dataflow), possibly but not necessarily, with the computing infrastructure based on the analog hardware. See the references for one specific viewpoint related to the subject.
TopConclusion
The major conclusion of this introduction to controlflow and dataflow supercomputing is that the best results are obtained with hybrid computers that synergize the two paradigms. The controlflow part is best used as the host, while the dataflow part is best used as the accelerator. A number of emerging applications are best implemented on hybrid computers.
Further advances in controlflow, dataflow, and hybrid computing need the creativity that is closely coupled with realities in the domains of underlying technologies and emerging applications. One study of creativity utilized in the related research so for could be found, together with related activities in the dissemination domain, in (Blagojevic2017, Bankovic2020).
However, the major challenge is to map efficiently the most sophisticated emerging algorithms and applications onto a properly synergized infrastructure that combines dataflow and controlflow paradigms.