Transparent synchronous dataflow: a functional paradigm for systems modelling and optimisation

Cheung, Wai Tak (2021). Transparent synchronous dataflow: a functional paradigm for systems modelling and optimisation. University of Birmingham. Ph.D.

Preview

Cheung2021PhD.pdf
Text - Accepted Version
Available under License Creative Commons Attribution Non-commercial.
Download (2MB) | Preview

Abstract

System modelling is the use of mathematical formalisms to model real world systems for the purpose of analysis, simulation and prediction. One of the most common ways to model a system is to create the dataflow among its various components. There are two main approaches on how dataflow graphs are constructed in these system modelling frameworks: ‘define-and-run’ vs ‘define-by-run’. The former approach first creates a dataflow graph and then executes it by pushing data into it. The latter however constructs the graph while computing with data on-the-fly. ‘Define-and-run’ is usually more efficient because many graph optimisations can be applied; ‘Define-by-run’ however handles dynamic models better. This thesis aims to develop a new functional paradigm for systems modelling and optimisation that exhibits properties of both approaches where dataflow graphs are dynamic but efficient.

We propose a new functional language, namely transparent synchronous dataflow (TSD), where dataflow graphs are constructed transparently with imperative commands to manipulate them explicitly; together with a synchronous mode of change propagation. The semantics of the language is designed on top of an unconventional graph abstract machine, Dynamic Geometry of Interaction Machine (DGoIM), which is natural for manipulating dataflow graphs. By using this semantics, the language is proved to be sound and efficient. Several experimental implementations were also created, including a native compiler for DGoIM and OCaml implementations for TSD.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):