Influence Control for Dynamic Reconfiguration of Data Flow Systems

Abstract

Influence control is a very challenging issue in dynamic reconfiguration and still not well addressed in the literature. This paper argues that dynamic reconfiguration influences system execution in four ways: functional update, functional side-effect, logical influence on performance and physical influence on performance. Methods including version control, transaction tracing, switching reconfiguration plan, and reconfiguration scheduling have been proposed for controlling the influence. These methods are integrated into the Reconfigurable Data Flow (RDF) model, which is designed to support the dynamic reconfiguration of stateless data flow systems. The RDF platform is an implementation of the RDF model on Java platform. The RDF platform is implemented as an open frame for different reconfiguration planning algorithms and scheduling policies to be simulated and their influence to be quantitatively compared on a single platform. Using the Data Encryption and Digital Signature System as a case study, tests have been done on the RDF platform to examine the influence of different reconfiguration planning algorithms and scheduling policies. Experimental results show that the methods proposed in this paper is effective in controlling the influence of dynamic reconfigurations.