10 likes | 149 Views
Evaluation. Problem. Analytical Using real topologies, real BGP data Calculate potential improvement Implement and emulate Code complete architecture Use software BGP speakers (GNU Zebra) Use Planetlab infrastructure Measure convergence, traffic balancing, overhead, scalability.
E N D
Evaluation Problem • Analytical • Using real topologies, real BGP data • Calculate potential improvement • Implement and emulate • Code complete architecture • Use software BGP speakers (GNU Zebra) • Use Planetlab infrastructure • Measure convergence, traffic balancing, overhead, scalability • Multihoming • Increasing trend • For failover and traffic load balancing • But BGP failover takes ~15 minutes! • Can only manage outgoing traffic • Data shows • Multihoming trends • Trying to manage incoming traffic can have bad global consequences • Goal • Reduce route failover time • Improve incoming load balancing • Reduce time to switch traffic between paths • Not worsen route flapping • Not worsen table bloat • Coexist with IGP & BGP • Incrementally deployable Updates by Prefix Multihoming OPCA Architecture Design Challenges • Local optimization is not enough • Locus of control is remote • Current BGP tricks are unscalable, ineffective • Global optimization is unattainable • Computationally complex • Link state scalability problem • Full disclosure of policies is bad • Middle ground needed • Logically separate routing control plane • Find loci of control • Negotiate policy control • Adapt to non-responsiveness • Adapt to network change OPP Messages Relationship Inferences AS Hierarchy AS Z MI PD E-BGP PA Other AS domains/networks in the Internet PA Directory RMAP AS X PA E-BGP sessions EBGP E-BGP PA MI PD AS W AS Y E-BGP speaker OPCA : Overlay Policy Control Architecture for BGP Sharad Agarwal, Chen-Nee Chuah, Randy Katz