Hora: Architecture-aware online failure prediction

摘要: Abstract Complex software systems experience failures at runtime even though a lot of effort is put into the development and operation. Reactive approaches detect these after they have occurred already caused serious consequences. In order to execute proactive actions, goal online failure prediction in advance by monitoring quality service or system events. Current look individual components as monolith without considering architecture system. They disregard fact that one component can propagate through cause problems other components. this paper, we propose hierarchical approach, called Hora , which combines predictors with architectural knowledge. The propagation modeled using Bayesian networks incorporate both results dependencies extracted from models. Our approach evaluated Netflix’s server-side distributed RSS reader application predict three representative types faults: memory leak, overload, sudden node crash. We compare monolithic show our improve area under ROC curve 9.9%.