Mapping concurrent applications onto architectural platforms

  • Authors:
  • Andrew Mihal;Kurt Keutzer

  • Affiliations:
  • University of California, Berkeley;University of California, Berkeley

  • Venue:
  • Networks on chip
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

Embedded system designers are faced with an expanding array of challenges 1in both application and architecture design. One challenge is the task of modelling heterogeneous concurrent applications. Another is the task of finding a programming model for heterogeneous multiprocessor architectural platforms. Compounding each of these challenges is the task of implementing heterogeneous applications on heterogeneous architectures. We believe that the key problem underlying each of these challenges is the modelling of concurrency, and the key to modelling concurrency is to capture concurrent communication formally in models of computation. This chapter broadly outlines a disciplined approach to the design and implementation of communication structures in embedded applications. Our approach combines the Network-on-Chip paradigm with the models of computation paradigm. We use models of computation to capture the communication requirements of an application as well as to abstract the capabilities of a communication architecture. Then, application requirements and architectural capabilities are matched using a discipline based on Network-on-Chip principles. In this chapter we describe this approach and present a case study where a Click network routing application is implemented on a multiprocessor architecture using this discipline.