Today's IP backbones are adequately provisioned to meet service level
agreements (SLAs) in terms of loss, delay, and port
availability. However, performance degradation is likely in the case
of failures (e.g., fiber cuts, router crashes), routing instability,
or router misconfigurations. This talk gives an overview of a
real-time monitoring infrastructure deployed within an operational
Tier-1 IP-backbone to analyze the characteristics of failure events.
The analysis is based on IS-IS routing updates from three different
points in the backbone. We will also investigate the potential impact
of failures on emerging services such as Voice-over-IP. In the second
part of the talk, we present results from a typical link failure
scenario in the backbone, including the re-convergence delay in
response to link failures and its effect on service disruption.  Our
results offer insights on two basic components for constructing a
realistic link failure model, and for defining "service" availability,
which we believe is the more appropriate basis for service-level
agreements to support emerging applications such as VoIP and VPNs.