Application of the Border Gateway Protocol in the Internet
obsoletes
- rfc-1655 — Application of the Border Gateway Protocol in the Internet
Extracted elements (22)
A border gateway should not advertise routes to external peers via an exit gateway B until all interior gateways within the AS can route traffic to B. This prevents black holes caused by BGP converging faster than the IGP.
Because a full AS path provides an efficient way to suppress routing loops and eliminates the count-to-infinity problem of distance vector algorithms, BGP imposes no topological restrictions on AS interconnections.
BGP-4 extends BGP-3 to support routing information aggregation and reduction based on CIDR, addressing the near-term Internet scaling crises: exhaustion of class-B address space and explosive routing table growth beyond manageable limits.
If all routers in an AS are BGP speakers (pervasive BGP), no BGP/IGP interaction or synchronization is needed. The IGP handles only intra-AS routing, and BGP routes may be advertised without waiting for IGP convergence because all routers already hold full BGP state.
Routing policies are not directly encoded in the BGP protocol; they are provided via configuration information. This allows BGP to enforce arbitrarily complex political, security, or economic constraints without changing the protocol itself.
Stub and multihomed AS's need not run BGP; they may use EGP or other protocols. This minimizes impact on the existing Internet infrastructure without degrading inter-AS routing quality for traffic to or from those AS's.
When operating BGP over Switched Virtual Circuits, eliminating periodic KEEPALIVE messages reduces SVC traffic costs. A circuit manager providing link-layer failure detection within a predictable finite period compensates for the absence of KEEPALIVEs.
When injecting BGP-4 NLRI into EGP2 or BGP-3, a potential information explosion can occur if an IP prefix from BGP-4 denotes a set of consecutive class A/B/C networks; injection of subnets requires the corresponding classful network to be injected instead.
When injecting BGP-4 routes carrying AS_SET path segments into BGP-3, the AS_SET segments shall be treated as AS_SEQUENCE, producing a single AS_SEQUENCE. This loses set/sequence distinction but preserves loop-suppression guarantees; however, policies based on AS_PATH ordering may be affected.
A conformant BGP-4 implementation is required to support four options for overlapping routes: install both less and more specific, install more specific only, install less specific only, or install neither.
A conformant implementation is required to support all of the following options when injecting BGP-4 routing information into BGP-3: inject default only, allow controlled deaggregation of specific routes, or allow export of only non-aggregated NLRI.
A route carrying the ATOMIC_AGGREGATE path attribute shall not be exported into BGP-3 or EGP2 unless the export can be accomplished without exploding the NLRI.
BGP implementations should allow an AS to control announcements of BGP-learned routes to adjacent AS's with at least per-address-prefix granularity and per-originating-or-adjacent-AS granularity. When a newly selected route cannot be announced to a peer that received the previous route, the system must explicitly withdraw the previous route.
BGP implementations should allow an AS to ignore routes containing certain AS's in the AS_PATH attribute by assigning them a weight of 'infinity'; the route selection process must then discard such routes.
BGP implementations should allow an AS to prefer a particular path to a destination. At minimum, an implementation shall support administratively assigning a degree of preference to a route based on the neighbor IP address, in the range 0 to 2^31 - 1.
Under the Tagged IGP method, a border gateway A must immediately propagate to all external peers any internal-peer-learned unreachability information, and must not propagate reachability information about prefix X to external peers until A has an IGP route with a tag matching the BGP tag for X.
Within a single autonomous system, BGP conversations among all BGP speakers must be uniformly either BGP-3 or BGP-4; a mixture is not permitted.
An AS is classified as one of three types based on transit behavior: a stub AS has a single connection and carries only local traffic; a multihomed AS has multiple connections but refuses transit traffic; a transit AS has multiple connections and carries both local and transit traffic.
BGP speakers in the same AS are internal peers; speakers in different AS's are external peers. External peers must share a common Data Link subnetwork. Using BGP across an intervening AS invalidates AS path information.
BGP SVC mode is selected by specifying Hold Time of zero in the OPEN message, causing the session to operate without periodic KEEPALIVE messages and relying on the circuit manager for link-layer reachability detection.
Local traffic originates or terminates within an AS; transit traffic does neither. Controlling transit traffic flow is stated as a major goal of BGP usage.
Security issues are not discussed in RFC 1772. The document explicitly states that security considerations are out of scope.