Large data center fabric tend to internal reasons as BGP routing protocols:
1.BGP than IGP has a simpler state machine and database;
2.BGP exchange routing, direct command, the algorithm does not need to run (such as SPF) to calculate the route ;
3.BGP hop traffic engineering support;
4. route for propagation, the smaller the range of the BGP (peer propagation, OSPF link state flooding required), and therefore more stable.
IBGP and EBGP selection:
1.EBGP session generally based on BGP peer link, IBGP generally based on the loopback interface of the BGP peers;
2.EBGP prevent loops using the AS-path, the same IBGP within the AS, use split horizon: the default route EBGP peers will be received from advertised to EBGP peers by IGBP and to other EBGP peers, the default route received from IBGP peers but not by giving other IBGP peers, since the principle of segmentation levels, generally use IBGP reflector (route reflector) or Alliance (use the confederation), generally using the RR;
3. Thus the EBGP generally used, simple and straightforward.
When the data center applications EBGP Fabric:
1. prevented by the AS-path attribute loop;
2. The default setting for the next-hop-self advertised route;
3. Other EBGP sessions automatically re-route learned from the advertisement;
of course, IBGP may also be used for the data center fabric, generally mimic the behavior EBGP:
1. set RR, and using the originator ID (originator-ID) and a list of clusters (cluster-list) to prevent loops;
2 is disposed next re-advertised route -hop-self.