T. Griffin and G. Wilfong, "Analysis of the MED Oscillation Problem in BGP", in IEEE International Conference on Network Protocols (ICNP), 2002.
|Analysis of the MED Oscillation Problem in BGP|
|Published:||IEEE International Conference on Network Protocols (ICNP), 2002|
|Abstract:||The Multi Exit Discriminator (MED) attribute of the Border Gateway Protocol (BGP) is widely used to implement "cold potato routing" between autonomous systems. However, the use of MED in practice has led to BGP persistent oscillation. The MED oscillation problem has been described with example configurations and complicated, step-by-step evaluation of dynamic route computations performed at multiple routers. Our work presents the first rigorous analysis of the MED oscillation problem. We employ the Stable Paths Problem (SPP) formalism that allows a static analysis of the interaction of routing policies. We give a formal definition of MED Induced Routing Anomalies (MIRA) and show that, in general, they can span multiple autonomous systems. However, if we assume that the BGP configurations between ASes follows a common model based on customer/provider and peer/peer relationships, then we show that the scope of any MIRA is always contained within a single autonomous system. Contrary to widely held assumptions, we show that a MIRA can occur even in a fully meshed IBGP configuration. We also show that a stable BGP routing may actually violate the stated semantics of the MED attribute.|
|Results:||Suggest a possible way of taming the MED attribute: having each AS write BGP routing policies that force the MED attribute to obey the rule of independent ranking. The rule of independent ranking states that a route's ranking should not depend on the (non)existence of other routes in the routing table. MED is unique in that it violates this rule.|