This talk will present our measurement and analysis of multiple origin AS (MOAS) conflicts in observed BGP updates, as well as a proposal for use of a community attribute and DNS to decrease risk in cases where MOAS conflicts may cause significant Internet routing disruption. Examples of such disruption include the AS8584 case in April 1998, and the instabilities generated by C&W peering changes in June of this year.

Seen in BGP routing table views, MOAS conflict is the case where a particular prefix originates from more than one AS. Using BGP routing tables from multiple views over 1279 continuous days, we analyzed total numbers of MOAS conflicts, duration of the conflicts, and relation to prefix length. We also classified the conflicts by the congruence of the multiple AS paths and analyzed the potential causes for the conflicts. These include intended uses to meet legitimate operational needs (multi-homing without BGP, private AS number substitution at egress, exchange points, anycast practices) and faults.

Both the total numbers of MOAS conflicts and the distribution of conflict duration suggest that MOAS due to faults represents a significant operational concern, even with the filtering lessons learned from AS8454 and similar events before and after. When a MOAS conflict occurs, we would like to enable routers to distinguish intended MOAS cases from route flapping or blackholing. Proposed solutions so far include the use of DNS (Bates/Bush1998) and uses of routing registries, including full-blown certifications (SBGP). In this talk we propose a much simpler and incrementally deployable approach using a new community attribute and DNS, which can provide adequate protection against faults from MOAS conflicts.

About the Presenter
Lixia Zhang has been involved in routing studies since the early days of IETF, as a member of the IETF Open Routing Group in the 80's. She is now working on a research project examining BGP fault tolerance enhancement.