Figure 1: IP address classes
This chapter is a précis of Chapters 18 and 20 of Doug Comer's book [Comer 2004] with additional material from Chapter 5 of Andy Tanenbaum's book [Tanenbaum 2003] . It explains the concepts of network addresses, network addressing, and IP datagrams.
The Internet is a large virtual network, an abstraction created in software, sitting on top of the underlying physical networks. It provides a uniform addressing scheme for all hosts which helps give the illusion of a large, seamless network. The IP standard specifies that each host is assigned a unique 32-bit IP address which is used for all communications with that host.
Conceptually, each 32-bit address is divided into two parts: a network prefix and a host suffix. Physical networks on the Internet have unique network addresses and all hosts attached to a particular physical network have unique host addresses.
The Internet includes arbitrary network technologies and both large and small physical networks. In order to accommodate this diversity, IP addresses are divided into classes, known as classful IP addressing, as illustrated in Figure 1.
The first four bits determine the class of the address. The first three classes, A, B, and C are called unicast addresses because they identify a single host. Class D is used for multicasting where a set of hosts share a common multicast address.
Classful IP addresses are called self-identifying because the class of the address can be computed from the address itself. The first four bits of the address can be extracted and used as an index into a table to determine the address class. This is illustrated in Figure 2.
Note than an address that begins with 1111 is
reserved and currently not used.
Although IP addresses are 32-bit numbers, humans use a dotted decimal notation which is easier to comprehend. Each 8-bit octet is written in decimal and the four octets have periods (dots) between them. Figure 3 illustrates some example 32-bit addresses and their equivalent dotted decimal forms.
The range of possible dotted decimal addresses goes from
0.0.0.0 through 255.255.255.255.
Dotted decimal notation works well with classful IP addresses because IP uses octet boundaries to separate the network prefix from the host suffix. The class must be recognised from the decimal value of the first octet. This is illustrated in Figure 4.
From Figure 4, it can be seen that the IP class scheme does not divide the 32-bit address space into equal size classes. This is illustrated in Figure 5, together with the maximum number of hosts per network.
The total number of hosts for each class also differ because of the number of bits required to identify the class; class A can contain 2,147,483,648 hosts, class B half of that, and class C half again.
Network numbers are managed by the Internet Corporation for Assigned Names and Numbers (ICANN) to avoid conflicts. Public networks are usually assigned a class C address unless they can justify a class B address. Class A is seldom justified.
Private networks (internets) select the appropriate class for their needs. Figure 6 illustrates a private internet containing one large network, two medium networks, and one small network.
The size of the cloud used to represent a physical network corresponds to the number of hosts expected on the network.
As the Internet grew, it became apparent that classful addressing was insufficiently flexible for certain situations. The solution to allow the boundary between the network prefix and the host suffix to occur on an arbitrary bit boundary is called subnet addressing or classless addressing.
Subnet addressing is implemented by storing a 32-bit address
and an additional 32-bit subnet mask
specifying where the boundary occurs. This subnet mask
consists of ones corresponding to the network prefix and
zeros corresponding to the host suffix. For humans, this
pair is written using Classless Inter-Domain
Routing (CIDR) notation. For example, the class
B network 128.10.0.0 would
be written as 128.10.0.0/16 to denote a 16-bit
network prefix. Another example is illustrated in Figure
7.
The network address 128.211.0.16/28 allows four
bits for host suffixes. A host suffix of all-zeros or
all-ones is special (see below), so only 14 hosts can exist.
An ISP can now allocate network addresses according to
demand. For example, the class B
network 128.211.0.0 could be subdivided into
128.211.0.16/28 for one customer and
128.211.0.32/28 for another. The ISP retains
most of the original address for other customers.
In addition to assigning a unique address to each host, the IP addressing scheme defines a set of special addresses that are reserved and never assigned to hosts. These addresses are illustrated in Figure 8.
The loopback address is usually
127.0.0.1 and can be used for more than
testing. It enables a user to run both a client and a
server on the same machine. Moreover, packets sent to the
loopback address never appear "on the wire", but are handled
internally by the IP software.
Each router on an internet will have two or more IP addresses assigned to it. This is because a router connects two or more networks together. Therefore, there will be a different IP network prefix for each physical network. This is illustrated in Figure 9.
Note that no hosts are shown connected to any of the networks. In reality, an IP address does not identify a specific host. Instead, it identifies a connection between a host and a network.
The designers of TCP/IP decided to make the fundamental delivery service, i.e. the network layer, a connectionless service with virtual packets created by a source host travelling from router to router until they reach a router that can deliver the packets.
Physical networks may use incompatible frame formats so, to overcome heterogeneity, IP software defines a virtual packet format that is independent of the underlying hardware. This packet is called an IP datagram, whose format is illustrated in Figure 10.
The header contains information that controls where and how the datagram is sent. The size of a datagram is determined by the application that uses it, thereby making IP adaptable to a variety of applications.
Datagrams traverse the Internet by following a path from the source, through routers, to the destination. Each router examines the destination address (in the header) and uses a routing table to determine the next hop. This is illustrated in Figure 11.
In (a), we have an internet with three routers connecting four physical networks. In (b), we have the conceptual routing table for router R2. Each entry in the table lists a destination network and the next hop to that network.
In practice, the routing table also needs a subnet mask associated with each destination network and the address of a router for each destination network. A practical example is illustrated in Figure 12.
In (a), we have four networks and three routers with the addresses assigned to each router interface. In (b), we have the routing table for the central router. In practice, most internets have more than four networks and a typical routing table would also contain a default route, as discussed above.
The router will first scan the routing table for an exact match, i.e. a 32-bit address; failing that, it will scan the table applying each subnet mask to the destination address in order to find a next hop router; failing that, it will use the default route, or cause an error if there is no default route.
Interestingly, the standard states that IP will make a best-effort attempt to deliver a datagram because there are a number of problems out of its control.
The underlying physical networks can cause such problems, so higher layers of protocol software are required to handle such these errors.
Figure 13 illustates the fields in the IP datagram header.
To keep the headers of most datagrams small, a variable-length list of options may also be present. Padding is required to expand the options to a 32-bit word boundary.
| Last modified: Thu Nov 24 10:40:11 2005 |