Experimental Assessment of Cloud Software Dependability Using Fault Injection

Part 5: Distributed ComputingInternational audienceIn modern cloud software systems, the complexity arising from feature interaction, geographical distribution, security and configurability requirements increases the likelihood of faults. Additional influencing factors are the impact of different execution environments as well as human operation or configuration errors. Assuming that any non-trivial cloud software system contains faults, robustness testing is needed to ensure that such faults are discovered as early as possible, and that the overall service is resilient and fault tolerant. To this end, fault injection is a means for disrupting the software in ways that uncover bugs and test the fault tolerance mechanisms. In this paper, we discuss how to experimentally assess software dependability in two steps. First, a model of the software is constructed from different runtime observations and configuration information. Second, this model is used to orchestrate fault injection experiments with the running software system in order to quantify dependability attributes such as service availability. We propose the architecture of a fault injection service within the OpenStack project