A Data-Driven Paradigm to Develop and Tune Data-Driven Realtime System

摘要

The authors have studied RESCUE (Realtime Execution System for CUE-series data-driven processors) which supports performance prediction, verification and tuning for realtime multi-processing system based on CUE processors. This paper proposes a data-driven paradigm totally supports development with considering both of user's requirements and engineering constraints. In the paradigm, a user defines functional specification and performance specification. The former describes at least algorithm and data-structure. The latter includes at least turnaround time and dataflow rate. The engineering constraints covers pipeline configuration and instruction set. In this paper, the authors firstly propose data-driven schema-based specification for turnaround time, dataflow rate and pipeline configuration. Implementation for realtime verification of turnaround time and processor load follows. This paper then proposes a scheme to localize emulation target, and discusses implementation to generate initial state and input stream needed for the localization. Lastly this paper preliminarily evaluates and shows interactiveness of RESCUE.