Optical Layer Failures in a Large Backbone.

ABSTRACTWe analyze optical layer outages in a large backbone, using data for over a year from thousands of optical channels carrying live IP layer traffic. Our analysis uncovers several findings that can help improve network management and routing. For instance, we find that optical links have a wide range of availabilities, which questions the common assumption in fault-tolerant routing designs that all links have equal failure probabilities. We also find that by monitoring changes in optical signal quality (not visible at IP layer), we can better predict (probabilistically) future outages. Our results suggest that backbone traffic engineering strategies should consider current and past optical layer performance and route computation should be based on the outage-risk profile of the underlying optical links.