OSPF is a routing protocol that is very well designed and is scalable to a very high degree. However, even though it does use areas very effectively to scale well, and it can use multiple processes, it is still not suitable for use on the Internet at large. Why? Well, for several reasons.
OSPF, like all IGPs, uses entries in the routing table that indicate the destination network and the next-hop IP. If it is designed well, this is fine if you have several thousand, or even tens of thousands of destination networks, and even thousands of OSPF routers. However, when you get to larger scales (hundreds of millions of destination networks, and millions of routers), the sizes of the routing tables and the processing power needed for routing decisions will quickly overwhelm the memory and CPU of any router, no matter how powerful.
For larger networks, a different type of routing process and algorithm is necessary, so that the routers involved will not be overwhelmed. That’s where comes in. BGP is a path-vector routing protocol. Instead of using next-hop addresses to get to where you want to go, it uses a hierarchical routing method. The whole internet is separated into Autonomous Systems (ASes), which are essentially groups of thousands of routers. BGP will maintain path information from AS to AS rather than from router to router, giving this hierarchical structure to routing that makes routing more efficient.
Links:
