REACT: a synthesis and evaluation tool for fault-tolerant multiprocessor architectures

A software testbed that performs automated life testing of a variety of multiprocessor architectures through simulated fault injection is discussed. It is being developed to meet the need for a generalized simulation tool which can evaluate system reliability and availability metrics while avoiding several of the limitations associated with combinatorial and Markov modeling. Incorporating detailed system, workload, and fault/error models, REACT can be more accurate and easier to use than many dependability prediction tools based on analytical approaches. The authors motivate the development of REACT, describe its features, and explain its use. Example applications for the software are provided, and its limitations are discussed