ARP is used by a layer 3 device to learn the layer 2 (MAC) address of a host for whom it only knows the IP address. For instance a router is connected to an Ethernet segment whose subnet address is 10.5.5.0/24; the router receives a packet destined for host 10.5.5.18 but it does not know the MAC address of that host. It broadcasts an ARP packet into the subnet and host .18 receives the packet and replies with its MAC address. The router then places that MAC address in its ARP cache next to the 10.5.5.18 IP address.
RARP (Reverse ARP) is used in the opposite way; it's how devices learn the layer 3 address of an end device for which it only knows the MAC address.
IARP (Inverse ARP) is used in frame-relay. Conceptually it's identical to RARP; i.e., it's for learning layer 3 addresses when only layer 2 addresses are known. In frame-relay's case the L2 addresses are DLCI's. A router knows it's got a FR circuit plugged into it; and it knows the DLCI for the circuit (either via static configuration or LMI-learned), but it does not know the IP address of the device at the other end of the circuit. And of course it needs to know this address so that when it gets traffic destined for it; it knows which DLCI to send the traffic onto. InvARP therefore sends a query onto the DLCI, the device at the other end responds with its own IP address.
I hope this also explains why IARP should be disabled if you're doing static mapping. If you're mapping destination IP addresses to your DLCI's on your own, you don't need InvARP (and it will also create needless traffic on the circuit).
Hope this helps; please rate this post if it does.
