BGP vs OSPF and IsIs

Question

Why hasn't BGP completely replaced OSPF and IsIs? What do the other two protocols handle that BGP does not already implement?

Answer 1

In what circumstances would it ever be practical for BGP to replace OSPF or ISIS?

BGP is an Exterior Gateway Protocol (EGP), it does not understand items like the bandwidth of links. Compare this to any Interior Gateway Protocol (IGP), including the two you mentioned, that make their routing decisions partly on available link speed.

BGP is more complex to configure properly than any IGP, add on the lack of support by lower end routers (not just bottom end routers) and the lack of automatic neighbour discovery and it becomes plain why BGP isn't about to take over any time soon, or ever in fact.

Answer 2

BGP is an Inter-domain protocol, which we use as example to assure communication between Autonomous System.
On the other hand OSPF is an Intra-domain protocol, which we use inside the AS in order to specify how routers communicate with each other.

Answer 3

Two additional observations that have not yet been covered in the previous answers:

1. OSPF and ISIS assume that they run within a single administrative domain, which means that all routers are run by the same organization. As such, these protocols share a lot of information (e.g. full topology) amongst the routers. BGP, on the other hand, is run between different administrative domains, i.e. between routers that are owned by different (possibly competing) organizations. As such, BGP is very careful to only share the minimum amount of information needed to make routing work (e.g. not the internal topology of an autonomous system). Also, instead of just focusing on finding the short path between two points, BGP provides a very rich "policy framework" that allows organizations to reflect business policies (= $$$) in BGP routing policies. For example, BGP provides the tools necessary to allow a company to decide whether or not they are willing to provide transit for another company. These BGP policy tools are very rich but also very complex, which makes BGP often an overkill as an Interior Gateway Protocol (IGP).

2. The current implementations of BGP typically scale to much larger networks (the entire Internet!) than the current implementations of OSPF and ISIS (typically hundreds of routers at the most). For that reason, some of the very largest networks run by the "hyperscale" companies (think Google, Microsoft, Amazon who have data centers with tens of thousands or even hundreds of thousands of servers) are sometimes using BGP instead of OSPF or ISIS inside their data centers. See RFC7938 (https://tools.ietf.org/html/rfc7938) for details. Alternatively, they can also run a centralized control plane using Software Defined Networking (SDN).

Answer 4

From CISCO's BGP page:

Q - Can IBGP be used in place of an IGP (RIP, IGRP, EIGRP, OSPF, or ISIS)?

A - Yes and no. Remember that the next-hop information from EBGP is carried into IBGP. If IBGP does not have a route to reach the next hop, then the route will be discarded. Typically an IGP needs to be used to exchange routes to the next hop, but this can be achieved by using static routes on all the routers running IBGP. So, the answer is yes if you want to use and maintain static routes. Otherwise, the answer is no.