The BIRD Route Server Software comes with the inherent limitation of being single threaded. While a gross of the deployments are by far too small (in terms of the number of peers/sessions and FIB entries) to reach any scalability limits, the application as a Route Server at a large IXP can easily overload a single BIRD instance.This talk presents a strategy to scale an existing deployment of BIRD on modern multi-core machines. The proposed solution spreads the load originating from peer-specific routing tables and an extensive amount of filtering to multiple instances of BIRD. Traditionally iBGP and BGP confederations are solution candidates, however they are not suitable to the IXP scenario. By also looking at the processing overhead and caveats of the configuration process we present the approach taken. We highlight current issues with BIRD deployments that have a number of peers in the hundreds and thousands range and also process a large number of prefixes.While our initial investigations are based on an unmodified version of the BIRD software we give an outlook towards further modifications that can help to make BIRD scale for possibly another ten years.
