GALS Partitioning by Behavioural Decoupling Expressed in Petri Nets
Danil Sokolov, Alex Yakovlev · 2014
Efficient design of a complex heterogeneous system requires detailed knowledge about the periodicity properties of its components and understanding the interaction patterns in their data exchange. Some of this information is usually available at the design time and facilitates basic optimisation through the insertion of sufficiently deep buffers between the communicating sub-systems. However, most of the information, e.g. the amount of data items to transfer or the relative difference in the iteration count of system components between communications, is data-dependent and can only be analysed dynamically at the run time. These data-dependent properties of the system still can be efficiently accounted for in globally asynchronous locally synchronous (GALS) design where the system is partitioned into independently clocked sub-systems that interact with each other in the asynchronous style. In this paper we introduce an approach to GALS partitioning based on the analysis of Petri net models of system components and the complexity analysis of their underlying algorithms.