Dataflow Weaknesses Analysis of Scientific Workflow Based on Fault Tree

Abstract

If potential contributors leading to system failure can be identified when a scientific workflow is modeled, a lot of system weaknesses may thus be revealed and improved. In this paper, we first identify a number of data dependency patterns in scientific workflows and their corresponding state functions. Then, a method to transform the state functions into fault tree symbols is presented. We use fault tree analysis method to identify critical elements and elements combinations that lead to the incorrect state of a final output and calculate the probability of the incorrect state of a final output based on the probabilities of the basic events in the analyzed workflow. Moreover, an importance measure is designed to prioritize the contributors leading to the incorrect state of a final output. Finally, the feasibility and effectiveness of the proposed methods are proved by example and experiments.