These are the basic differences.
With Fast Switching, the forward information database (fib) is built gradually on an as needed basis. It starts empty and is built as users need to find destinations.
With the first packet a user sends to destination $x, the ip routing table (rib) is consulted to find the outbound interface toward the ip address of the destination, the arp table is consulted to find the mac addy of the next hop. Being that looking up the destination via the ip routing table/rib is done in software (process switched), it takes more resources to perform this action.
After the lookup, the destination ip and outbound interface is sent to the fib so all subsequent packets to this destination do not have to be looked up (per ip routing table/rib).
With CEF, the full RIB is pushed/mirrored to the CEF FIB after boot so the router/switch does not have to gradually build the CEF FIB. Also MAC addresses for the next hop are contained in the Adjacency table. There is a hash table with pointers from the FIB to the Adjacency Table. This makes it easier for this entire process of packet switching to be programmed into hardware. Subsequently, the first packet sent to a destination does not have to be sent to software unless particular security related features are enabled. In the case of 6500 switches with linecards containing DFC/Distributed Forwarding Cards, each linecard will have a copy of the CEF FIB and Adjacency Table so the “packet switching” can be done on the linecard.
