It should be clear from our discussion thus far that mobility management consists largely of routing data packets (NPDUs) to hosts which change location and network access frequently relative to conventional hosts.1.10 By definition, routing is a network layer function and thus mobility management should also be a network layer function. Efficient support for mobility must be designed into the network layer for any mobile WAN.
As observed by [IOAN93]: "The problem of Mobile Internetworking [can be] posed as follows: how to provide seamless and transparent network connectivity to mobile networked computers (or other communicating devices) as they change location, networking interfaces, or even service providers. We term this work ÔMobile Internetworking' because it enables mobile entities to communicate within an internetwork, i.e., a network of networks, and not just a local, connected network."
Data packets are routed across conventional internetworks via their destination host network layer addresses. The ability to route packets toward their final destination is based on the fact that network layer addresses are typically composed of a network-identifying part and a host-identifying part. The network-identifying part of the address specifies on which network the host may be found. The host-identifying part of the address specifies which host on the network is desired.
Figure 1.6 depicts an IP address, which is four bytes long. The network-identifying part is called the netid and the host-identifying part is called the hostid. Originally the separation between these fields was required to be at one of the three byte boundaries; three corresponding classes of network address space were defined. Now with IP version 4 (IPv4), the boundary between netid and hostid is identified via a bit mask, allowing complete flexibility for IP address space allocation.
Because of the rapid adoption of IP-based technology and the Internet, IP address space had become a precious resource by the early 1990's. Four bytes sounds like a lot of addresses, but typically host address assignments are relatively sparse based on organizational and network boundaries.1.11 Partly as a result of the address space limitation and also because of the protocols defined, the IP addresses are typically manually administered and statically assigned to hosts. IPv6 (the "next generation") [BRAD96] has extended the size of IP addresses to 16 bytes, largely to alleviate these concerns about address space availability.
The assignments of network layer addresses to hosts is based on the topology (state of connectivity) of the network. Routing information (contained in the routers in the network) can be considered to be a form of distributed database, where a partial view of the network topology information is contained in each router. Each router must be capable of selecting the "next hop" for each packet based on its ultimate network layer destination address.
Figure 1.7 depicts routing in conventional data networks. A data packet is forwarded from its source located somewhere in the rest of the world to its destination host, Host 1. Table 1.1 depicts the corresponding entries in the routing tables at Routers W, X, Y and Z, which participate in the packet forwarding. The null entry in the "next hop" field for Router X indicates that Host 1 is local to Router X.
Over time this distributed routing database evolves to reflect changing WAN connectivity. As links and nodes come and go, the routing tables must reflect these changes. This routing table adaptation can either be manually administered (called "static routes") or automatic (i.e., based on router protocols such as RIP and OSPF) [PERL92]. Automatic routing table updates support changing network environments due to configuration changes (intentional!) and failures (unintentional!).
Novell's proprietary IPX protocol uses ten-byte network layer addresses in a creative fashion to support a plug-and-play network configuration capability. The first four bytes of the address represent the network and are called "network." The last six bytes are called "node" and identify the host ("node" in Novell terminology). This field is identical to the device's permanent six-byte MAC sublayer identifier.
In IPX there is no requirement for network administrators to explicitly configure each node address, only each router's network value. Plug-and-play capability-certainly a factor in mobility-is supported in that a host only needs to determine its "network" value by querying a local router when the host is joined to a network. Then the node must notify each of its application peers of its new "permanent" address which it has self-configured.
This self-configuration by IPX significantly reduces the effort required to move devices from one location to another and prevents node address conflicts. It also eliminates the need for a protocol, such as ARP1.12 , to provide the network layer to data link layer address mapping for local routing.
Rapid convergence of router protocols (i.e., adaptation of the routing information database to changing network state) is a primary concern in WANs. Routing protocols must converge more quickly than network topology changes occur or the internetwork operation will break down from the congestion caused by misdirected packets. In fact, one of the biggest drawbacks to the popular RIP protocol is its slow convergence in large-scale internetworks.
Conventional data network routing table updates must be done frequently enough to prevent congestion in the event of a link or router failure or network reconfiguration. [IOAN93] observes that "if the links go up and down faster than the [routing] protocols can converge, routing may not be possible even though the physical paths exist."
The need for rapid convergence is amplified in mobile environments, where the movement of hosts creates and destroys links (to those hosts) dynamically and presumably more frequently than failures occur. Routing information needs to be rapidly shared amongst mobility routers to ensure consistency between their routing tables and the represented physical network topology.
[KRIS95] notes that "conventional routing protocols were not designed for networks where the topological connectivity is subject to frequent, unpredictable change." Although current routing information protocols support adaptive routing updates to reflect network and host connectivity changes, these protocols are designed for infrequent updates and failures, and thus inadequate to support mobility.