Authors:
Salvador Fallorina
;
Paul Thienphrapa
;
Rodrigo Luna
;
Vu Khuong
;
Helen Boussalis
;
Charles Liu
;
Jane Dong
;
Khosrow Rad
and
Wing Ho
Affiliation:
California State University, Los Angeles, United States
Keyword(s):
Distributed data flow, pipelined task mapping, scheduling, fault-tolerance, real-time decentralized control.
Related
Ontology
Subjects/Areas/Topics:
Computer and Microprocessor-Based Control
;
Informatics in Control, Automation and Robotics
;
Real-Time Systems Control
;
Signal Processing, Sensors, Systems Modeling and Control
Abstract:
Complex control-oriented structures are inherently multiple input, multiple output systems whose complexities increase significantly with each additional parameter. When precision performance in both space and time is required, these types of applications can be described as real-time systems that demand substantial amounts of computational power in order to function properly. The failure of a subsystem can be viewed as the extreme case of a non-real-time response, so the ability of a system to recognize and recover from faults, and continue operating in at least some degraded mode, is of crucial importance. Furthermore, the issue of fault-tolerance naturally arises because real-time control systems are often placed in mission-critical contexts. Decentralized control techniques, in which multiple lower-order controllers replace a monolithic controller, provide a framework for embedded parallel computing to facilitate the fault-tolerance and high performance of a sophisticated control
system.
This paper introduces a fault-tolerant concept to the handling of data flows in multiprocessor environments that are reminiscent of control systems. The design is described in detail and compared against a typical master-slave configuration. A distributed data flow architecture embraces tolerance to processor failures while satisfying real-time constraints, justifying its use over conventional methods. Both master-slave and distributed data flow designs have been studied with regards to a physical control-intensive system; the conclusions indicate a sound design and encourage the further division of computational responsibilities in order to promote fault-tolerance in embedded control processing systems.
(More)