Reliable Distributed Computing for Decision Support Systems

Abstract

Real-time decision support typically requires that several tasks such as on-line monitoring, simulation, prediction, state estimation, and real-time control are run concurrently. This is implemented most effectively in a true distributed system rather than in a time-shared uni-processor environment. Moreover, the physical size of many decision-support systems such as water distribution networks, involving hundreds or even thousands of pipes and network nodes spread over a large geographical area, calls for a control of computational complexity of the constituent decision support tasks. This is usually achieved by topological decomposition of systems and parallelisation of the respective algorithms1,8. The parallel tasks can then be executed on networked workstations which provide a powerful and flexible (upgradable according to specific requirements) computing environment. The performance of such systems has been recently significantly enhanced by the use of high-speed communications links such as FDDI, ATM, Gigabit Ethernet etc 2.