Dependability evaluation of distributed systems through partitioning fault injection

Fault injection intents to insert artificial faults in a target system under test to assess their behavior in abnormal situations. However, few efforts focus on partitioning faults that affect distributed applications. The difficulties in reproducing network partitionings remain open. This paper presents a novel tool that emulate partitioning in a real IP network to evaluate the dependability of distributed applications. The tool is composed of distributed instances of a communication fault injector and an experiment coordinator. Using the tool, partitioning scenarios can be described at a high level of abstraction allowing to create high expressive faultloads and realize effective fault injection experiments.