Routing

These questions, for the most part, relate to routing.

Q: Volume I discusses proxy ARP in which one router answers ARP requests on behalf of others (p.142). Can Proxy ARP be used across more than two physical networks? If a router R is responding (falsely) to one (phony) network prefix, why can it not respond to k-1 such prefixes?

A: It seems that you are asking about a single computer with connections to K+1 networks. Can such a computer act as a proxy for all of them? That is, when an ARP request arrives over one network, can the request be for a host on *any* of the other networks? Sure, such a scheme is possible. The computer connected to the networks will need a database that specifies which computer attaches to which network. The database is used for two purposes: for proxy ARP and for forwarding datagrams when they arrive.

Q: Subnet addressing mentioned in an exercise. (I found information in volume I, p.150.) Isn't it true that the paths from the same source to two different destination hosts residing on different subnets of the same network will be identical except for the final binding? That is, why store subnet masks for next hops of distant physical networks -- aren't the next hops the same?

A: The idea of subnetting is to use a single IP network prefix for multiple physical networks (to conserve prefixes). Subnetting only affects hosts and routers at a given site -- everywhere else in the Internet, the addresses are treated like normal IP addresses. Thus, all datagrams headed to any of the subnetted networks follow the same path through the Internet. Only when they arrive at the site using subnetting do they follow paths to one of the individual networks in the subnetted group. The point is that routers do *not* need to store subnet masks for networks that are far away from a destination that uses subnet addressing.

However.... having said all that, I should also tell you that managers sometimes install subnet-specific routes for administrative reasons. For example, suppose that the engineering department in a large corporation has two subnets of a Class C network number, with one subnet being used for engineers' workstations and the other being used for equipment under beta test. The network manager of the accounting department may choose to route packets headed to the workstation subnet along a more secure path than packets headed to the testing subnet (e.g., to make it possible to send accounting information).

Q: More on subnet routing (volume I, p. 150). Do *all* routers contain an extra field for subnet addressing or just local routers physically attached to networks partitioned into subnetworks?

A: Commercial routers all have the capability to do subnet routing. In fact, a single mechanism (the 32-bit mask) can be used to allow routers to specify a network, a subnet, or a host. Thus, it is *possible* for a manager to specify a unique route for each host on a network. In practice, one seldom uses host-specific routes, but they can come in handy when there are "exceptions." For example, one can direct packets headed to the CEO's computer along a completely different path than those headed for computers of other employees.

Q: With regard to CIDR routing, also mentioned in an exercise. (I found information in volume I, p.155.) How do distant routers handle datagrams destined for CIDR networks?

A: That's simple because supernetting, like subnetting, is local to a given site. In fact, the site is merely assigned a set of contiguous class C prefixes in place of a single higher class prefix. For example, the site might be assigned 256 class C prefixes instead of a single class B prefix. If a router does not understand CIDR aggregation, the router will have 256 individual routes in it for the 256 class C numbers (the routes will all point to the same next hop). If the router does understand CIDR, it will have a single entry that gives the range of class C numbers and the corresponding next hop. Thus, CIDR only saves space in router tables; it does not make a significant change in the way addresses are assigned.

Q: It seems that it's possible that all (core) routers must be able to handle (1) standard IP addresses, (2) subnet addresses, and (3) supernet addresses. Do the address-handling capabilities of routers along the way indirectly dictate the paths of traveling datagrams -- especially does it matter whether routers possess a CIDR capability?

A: No. Both supernetting (CIDR) and subnetting are "backward compatible" with the original addressing scheme. Except for the local site using one scheme or the other, routers need not know about them. (Knowing about CIDR does permit routing tables to be smaller, but that only optimizes the performance or the cost; datagrams follow the same routes in any case).

Q: Can a user's computer (i.e., a host) have connections to two different networks?

A: Yes. Such computers are called *multihomed hosts*. Although such arrangements are possible, they are rare.

Q: What's the difference between a multihomed host and a router?

A: A multihomed host does not forward packets from one network to the other, and it does not participate actively in routing update protocols (i.e., it doesn't advertise that it has multiple connections).

Q: Can a multihomed host be a member of two subnets of a subnetted network?

A: In theory, sure. It has two subnet masks, and handles each subnet independently.

Q: Can a multihomed host be a member of a subnetted network and a supernetted network?

A: Sure. The host simply thinks it has two network connections, one of which uses subnetting and the other which is not subnetted.

Q: Does the term 'point-to-point' necessarily refer to *serial* line communication?

A: A point-to-point network is any system that connects from one computer directly to another (as opposed to a shared media network that interconnects multiple computers). Many point-to-point networks consist of a serial data circuit leased from the phone company. It is also possible to connect two computers using a parallel point-to-point network, but that is less common.

Q: Does the term 'intranet' refer to an internet formed from point-to-point serial links?

A: No. An intranet is simply a private network -- a set of networks and routers owned by one organization. A company might choose to use serial links to connect its routers, but that would be unusual. Most companies use shared media networks instead (e.g., Ethernet or FDDI).

Q: When two routers use a routing protocol to exchange routing information, must they share a common network?

A: Yes. However, you must remember that in the IP world, a "network" can be a point-to-point leased line. Thus, if a corporation has a pair of routers, with one router at site A and another at site B, the corporation only needs to lease a dedicated line between them in order to run a routing protocol.

Q: If a point-to-point connection between two routers is a "network," does it need an IP prefix?

A: Although a point-to-point connection is a network and *can* be assigned an IP prefix, doing so is a waste of (scarce) IP addresses. Thus, most managers do not assign IP prefixes to point-to-point lines. Instead, they use a technique known as anonymous connections that does not assign an IP prefix to a point-to-point line. All the routing across an anonymous connection is performed without using the next-hop address (only the interface number in the routing entry is used).