By design, IBGP peers do not advertise iBGP routes to other iBGP peers. In order for iBGP peers to learn all the routes within the autonomous system as well as all the external routes, the iBGP peers would have to be fully meshed. This means for n iBGP peers there would have to be n*(n-1)/2 iBGP sessions. In a large autonomous system network configuration would become an issue.
Route Reflection is one of the alternate solutions to alleviate this problem. In the BGP network, one of the iBGP speakers is designated as the route reflector. The route reflector advertises the routes it learns to other iBGP peers.
In a route reflector configuration the other iBGP peers are classified as clientpeers and non-client peers.
The action taken by the route reflector (after determining the best route) depends on whether the best route was received from a client peer or a non-client peer. If the route was received from a client peer, the route reflector will reflect that route to all the client peers and non-client peers.
If the route was received from a non-client peer, then the route is advertised to all its configured clients.
Route reflection introduces two new discretionary attributes: Originator ID and Cluster List, which are used in determining the best path as defined in BGP route selection.
In an Autonomous System more than one route reflector can be configured.
When a BGP speaker shuts down, planned or unplanned, the routes that are advertised by the speaker and reachable via the speaker now become unreachable. Upon detecting that the BGP speaker has restarted, the peers delete the routes and re-add them when the restarting router advertises them again. This results in route-flap across the BGP connectivity and impacts multiple routing domains causing transient instability in the network.
The Graceful Restart capability is supported as a 'helper router` on the HP 3500, 5400, and 8200 product series. In 'helper only' mode the router helps the other restarting router by holding the received routes from it as stale routes and not dropping them.
On the HP 8200 product series, the Graceful Restart capability is supported as a restarting router in non-stop routing mode.
-
To establish a BGP session with a peer, a BGP GR Restarter sends an OPEN message with GR capability to the peer.
-
Upon receipt of this message, the peer is aware that the sending router is capable of Graceful Restart, and sends an OPEN message with GR Capability to the GR Restarter to establish a GR session. If neither party has the GR capability, the session established between them will not be GR capable.
-
The GR session between the GR Restarter and its peer goes down when the GR Restarter restarts BGP. The GR capable peer will mark all routes associated with the GR Restarter as stale. However, during the configured GR Time, it still uses these routes for packet forwarding.
-
After the restart, the GR Restarter will reestablish a GR session with its peer and send a new GR message notifying the completion of restart. Routing information is exchanged between them for the GR Restarter to create a new routing table and forwarding table with stale routing information removed. Then the BGP routing convergence is complete.
When the inbound policy-filter for a peer changes, the routes advertised by the peer must be presented to the policy-filter engine to take effect. This means that all the routes that were received from a peer will have to be preserved in the router and this would raise the demand on memory and CPU resources of the router. The route refresh capability allows the router to request the peer to re-advertise the routes thereby avoiding the requirement to keep a copy of all the routes that were received from all the peers.