IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 15, NO. 2, APRIL 2007 253
Network-Wide Prediction of BGP Routes
Nick Feamster and Jennifer Rexford, Senior Member, IEEE
Abstract—This paper presents provably correct algorithms for
computing the outcome of the BGP route-selection process for each
router in a network, without simulating the complex details of BGP
message passing. The algorithms require only static inputs that can
be easily obtained from the routers: the BGP routes learned from
neighboring domains, the import policies configured on the BGP
sessions, and the internal topology. Solving the problem would be
easy if the route-selection process were deterministic and every
router received all candidate BGP routes. However, two impor-
tant features of BGP—the Multiple Exit Discriminator (MED) at-
tribute and route reflectors—violate these properties. After pre-
senting a simple route-prediction algorithm for networks that do
not use these features, we present algorithms that capture the ef-
fects of the MED attribute and route reflectors in isolation. Then,
we explain why the interaction between these two features pre-
cludes efficient route prediction. These two features also create dif-
ficulties for the operation of BGP itself, leading us to suggest im-
provements to BGP that achieve the same goals as MED and route
reflection without introducing the negative side effects.
Index Terms—Networks, protocols, routing.
I. INTRODUCTION
TO CONTROL the flow of traffic through their networks,operators need to know how configuration changes affect
the routes that each router in the network selects. The outcome
of the route-selection process depends on the routes advertised
by neighboring domains, the internal topology, the interdomain
routing policies, and the intradomain link weights. Ordinarily,
computing the outcome would require a complex simulation
of routing-protocol dynamics. Instead, we present efficient al-
gorithms that compute the outcome of the BGP route-selec-
tion process without backtracking. In designing our algorithms,
we grapple with two features of the Border Gateway Protocol
(BGP) [1]: limited visibility into the available routes for each
destination and non-deterministic ranking of these routes.
A. Backbone Network Engineering
The flow of traffic through a backbone network depends on
the interactions between three routing protocols, as shown in
Fig. 1.
External BGP (eBGP): Routers in the AS use eBGP to
receive route advertisements from neighboring ASes. For ex-
ample, the routers , and each have eBGP sessions with
neighboring ASes. The routers may apply an import policy to
modify the attributes of the routes learned from the neighbor,
Manuscript received August 27, 2004; revised November 4, 2005; approved
by IEEE/ACM TRANSACTIONS ON NETWORKING Editor S. Seshan.
N. Feamster is with the College of Computing, Georgia Institute of Tech-
nology, Atlanta, GA 30332-0180 USA (e-mail: feamster@cc.gatech.edu).
J. Rexford is with the Computer Science Department, Princeton University,
Princeton, NJ 08540 USA (e-mail: jrex@cs.princeton.edu).
Digital Object Identifier 10.1109/TNET.2007.892876
Fig. 1. Network with three egress routers connecting to two neighboring ASes.
Solid lines correspond to physical links (internal links are annotated with IGP
link weights) and dashed lines correspond to BGP sessions. Thick lines illustrate
the shortest path from one router to its closest egress point for reaching the
destination.
with the goal of influencing the selection process in Table I that
each router applies to select a single best BGP route for each
destination prefix.
Internal BGP (iBGP): The routers use iBGP to disseminate
the routes to the rest of the network. In the simplest case, each
router has an iBGP session with every eBGP-speaking router,
forming an “full mesh” configuration. If two routes are equally
good through the first four steps in Table I, the router favors an
eBGP-learned route over an iBGP-learned one. In Fig. 1, router
receives three iBGP routes, from routers , and . Upon
learning routes with the same local preference, AS path length,
origin type, and MED values, router uses the IGP to break
ties between the remaining routes.
Interior Gateway Protocol (IGP): The routers run an Interior
Gateway Protocol (IGP) to learn how to reach each other. Two
common IGPs today are OSPF [2] and IS-IS [3], which compute
shortest paths based on configurable link weights. The routers
also use the IGP path costs in the sixth step of the BGP route-
selection process in Table I. In Fig. 1, router selects the route
with the smallest IGP path cost of 2, learned from router .1
After selecting a route to each destination, each router com-
bines the BGP and IGP information to construct a forwarding
table that maps the destination prefix to the outgoing link along
the shortest path. In Fig. 1, the forwarding path consists of the
thick lines from the ingress link at router to the egress link at
router .
1If two routes have the same IGP path cost, the router performs an arbitrary
tiebreak in the seventh step in Table I.
1063-6692/$25.00 © 2007 IEEE