International audienceSystem-level Mean Time Between Failures (MTBF) is usually evaluated using individual component-level reliability metrics. System-level failures are categorized by Reliability, Availability and Serviceability (RAS) metrics. However, RAS evaluation at the system-level requires precise mapping between component failure modes, their system failure signatures and system reliability requirements. In this paper, RAS analysis carried out on internet switches in a top-down hierarchical fashion is presented. Results show availability of failure classification at a lower-level of design allows for better fault management and improved RAS metrics at the system-level. A hierarchical modeling format is proposed to standardize the reporting of component failure modes to improve the system level modeling of RAS