Hi there,

First of all about how the provider assigns the DLCI to the customer --> The DLCI is local significant (significant per each local loop - per each site and its opposite FR Switch).

The local DLCI significance simply means that for a customer having an end-to-end PVC between 2 routers in 2 different site -> There can be 2 DLCIs -> One from the first CE router to the facing FR switch in the provider network and the another DLCI for the other CE router with its facing FR Switch --> Cross-connections are configured within the provider?s network to establish this end-to-end PVC for the customer.

Second regarding the InARP, as a simple definition it is used to obtain the IP address using the DLCI (on the LAN it is used to obtain the IP address using the MAC address), how the InARP depends upon the topology as follows:

Point-to-point --> Doesn?t need either InARP or Static Mapping --> point-to-point subinterfaces ignore the received InARP information --> to send the packet over the only possible DLCI on that point-to-point subinterface --> ?send all packets for addresses in this subnet using the only DLCI on the subinterface? logic --> Addresses of remote destinations for each point-to-point subinterface are not resolved since they are point-to-point and traffic must be sent to the peer at the other end.

Multipoint / Physical --> Requires either InARP or Static mapping using the ?frame-relay map? command, and in most cases InARP does not help, because InARP messages only flow across a VC, and are not forwarded (most scenarios should be partial mesh or hub-and-spoke, or any other scenario where not all the sites are directly connected); The solution is to add a frame-relay map command to the router configuration which disables Inverse ARP by default: ?frame-relay map ip broadcast?.

Hub-and-Spoke --> The Hub can count on InverseARP, but spokes can't use InverseARP as InverseARP is point-to-point (only flow accross a VC) and can't go from a spoke to a spoke through a hub (InARP messages only flow across a VC, and are not forwarded) --> Spokes must have the static mapping (if using Multipoint / Physical interface - While remember that if using point-to-point there is no need for either InARP or Static Mapping) --> "frame-relay map ip [subnet]"

NOTE: Point-to-point subinterfaces can be unnumbered for use with IP, reducing the addressing burden that might otherwise result.

HTH, Please rate if this helps,

Mohammed Mahmoud.