Chapter 18 -- IP: internet protocol addresses

(rev. 01/11/08)

Notes On Chapter Eighteen -- IP: Internet Protocol Addresses

18.1 Introduction
- Here we describe the protocol software that transforms a heterogeneous set of computers into an internet -- one virtual network with a uniform interface. In this chapter, we start with a discussion of the addressing scheme.
18.2 Addresses For The Virtual Internet
- The Internet is built "on top of" whatever the network implementation is at the lower levels. TCP/IP is designed to be adaptable to any underlying network implementation.
- IP addressing must be a uniform standard -- across all member nets -- independent of lower level addressing schemes.
- Internet routers switch IP packets based on the IP address.
18.3 The IP Addressing Scheme
- IP addresses are 4-byte, 32-bit numbers.
- Every IP packet contains a source IP address and a destination IP address.
18.4 The IP Address Hierarchy
- Some initial part of an IP address (a prefix) identifies a particular network. The rest of the address (the suffix) identifies a particular host interface on that network.
- Two different host interfaces must never be assigned the same IP number.
18.5 Original Classes Of IP Addresses
- The Internet is comprised of a large number of networks, some containing a large number of hosts, and some containing smaller numbers of hosts.
- Three of the original classes of IP address were designed to accommodate networks of different sizes.
  - Class A addresses are for big networks with many, many hosts.
    - The first bit of a class A address is "0".
    - That means that the first "octet" is 127 or less.
    - In a class A address, the first octet is the network address and the other 3 octets are used for host addresses.
    - That means there can be at most 2 to the 7th (128) class A networks and almost 2 to the 24th hosts in each network. That's about 16 million!
    - Example 18.71.0.151 (strawb.mit.edu) is a class A address. (18 is the network prefix.)
  - Class B addresses are for medium sized networks.
    - The first two bits of a class B address are "10".
    - That means that the first "octet" is between 128 and 191 (inclusive).
    - In a class B address, the first two octets are the network address and the other 2 octets are used for host addresses.
    - That means there can be at most 2 to the 14th (16K) class B networks and almost 2 to the 16th (64K) hosts per network.
    - Example: 130.17.70.165 (ishi.csustan.edu) is a class B address. (130.17 is the network prefix.)
  - Class C addresses are for small networks.
    - The first three bits of a class C address are "110".
    - That means that the first "octet" is between 192 and 223 (inclusive).
    - In a class C address, the first three octets are the network address and the last octet is used for host addresses.
    - That means there can be at most 2 to the 21st (2Meg) class C networks and almost 2 to the 8th (256) hosts per network.
- There is a class of IP addresses just for multicasting.
  - A multicast address starts "1110" so the first octet is between 224 and 239 (inclusive)
- Addresses that start "1111" are reserved. They have first octet between 240 and 255 (inclusive).
18.6 Computing The Class Of An Address
- The so-called classful IP addresses are self identifying. A router can very quickly determine the class of an address by looking at the first few bits.
18.7 Dotted Decimal Notation
- Humans usually write IP numbers in "dotted decimal" notation like "129.52.6.0".
- Each decimal part stands for the value of one of the octets.
- Note that the decimal parts have to be in the range 0..255 (inclusive).
18.8 Classes And Dotted Decimal Notation
- As explained in section 18.5 of these notes, the class of an address determines the possible range of values of the first octet or dotted decimal part of the address.
18.9 Division Of The Address Space
- As shown in section 18.5 of these notes, the class of an address determines line of division between the network and host parts of the address. This in turn determines how many numbers exist for class A, B, and C networks. (128, 16384, and 2,097,152).
18.10 Authority For Addresses
- The Internet Assigned Numbers Authority (IANA) is the central authority that ensures that IP network numbers are assigned uniquely in the Internet.
- IANA delegates authority to ISP's and various groups to assign numbers.
- The owners of a private internet can assign whatever numbers they like. (However it is wise to do so in a manner that will allow the private internet to connect to the (public) Internet in the future.)
- RFC 1597 has guidelines for numbering in private internets.
18.11 A Classful Addressing Example
- One chooses a class A, B, or C network address based on the number of hosts that will eventually exist in the network.
- Within a network the prefix of the IP address of each host is the same.
- The suffix of each host in the network is different from that of all the other hosts.
- Choosing a network class is problematic because the only choices give capacity for ~255, ~64,000, or ~16,000,000 hosts!
18.12 Subnet and Classless Addressing
- In classful addressing, prefix that denotes the network part of the address must have a length of one, two, or three octets -- i.e. eight, sixteen, or twenty-four bits.
- The idea of subnet and classless addressing is to get more flexibility by allowing network prefixes to have other lengths.
18.13 Address Masks
- There is no big trick to making classless addressing work, but one does need to know where the boundary is between the prefix and the suffix. If you know the address is classful, then you can tell where the boundary is by looking at the first few bits of the address. (There is a predetermined rule.)
- With classless addressing, no "rule" tells you where the boundary is. The information is usually conveyed by using a 32-bit network mask. In the mask the bits corresponding to the prefix are all ones and the other bits are zeros.
- If D and (A,M) are respectively 32-bit destination address, network address, and the appropriate network mask for address D, then A==(D&M) if and only if D is a host in network A.
- Routers use the idea above to figure out to which network to route packets. When (D&M) matches a network address A in a routing table, it means the packet with address D must be delivered to network A. The other information in the routing table row containing A will tell the software where to send the packet for its next hop.
  
  Example:
```
  A == 1000 0000  0000 1010  0000 0000  0000 0000  (128.10.0.0)
  D == 1000 0000  0000 1010  0000 0010  0000 0011  (128.10.2.3)
  M == 1111 1111  1111 1111  0000 0000  0000 0000  (255.255.0.0)
D&M == 1000 0000  0000 1010  0000 0000  0000 0000  (128.10.0.0)
```
  Note that D&M == A.
18.14 (Classless Inter-Domain Routing) CIDR Notation
- An "address" of this form: 128.10.43.134/20 denotes that the first 20 bits are the network prefix.
  
  So here, the mask would be
```
M == 1111 1111  1111 1111  1111 0000  0000 0000  (255.255.240.0)
```
18.15 A CIDR Address Block Example
- 128.211.0.16/28 is an address with a 28-bit network prefix. The address is:
```
128.211.0. 0001 0000
```
  (It must be understood that the first three octets are written in dotted-decimal notation and the last octet is expanded out in binary.) The 28-bit network prefix is:
```
128.211.0. 0001
```
  and the 28-bit mask is:
```
255.255.255. 1111 0000
```
- 128.211.0.32/28 is another address with a 28-bit network prefix. The address is:
```
128.211.0. 0010 0000
```
  (Note that the two addresses have different prefixes. Therefore it is clear that the two addresses can (and must) be assigned to different networks.)
- If an ISP assigns 128.211.0.16/28 and 128.211.0.32/28 as network addresses, then each network can use its four bits of host suffix to assign 14 host numbers. (The all-ones suffix and the all-zeros suffix are reserved for special purposes, hence 14 hosts instead of 16.)
- If an ISP is given the class B network address 128.211.0.0, it can use the CIDR scheme to break that address up into addresses like the ones illustrated above, and many additional networks of different sizes, by making appropriate choices for the sizes and values of address prefixes.
- As another example of what can be done, the ISP could could assign the network address 128.211.0.128/26, which is:
```
128.211.0. 1000 0000
```
  Here the mask would be:
```
255.255.255. 1100 0000
```
  This would leave 6 bits for host numbers, and so there would be a possibility of 62 hosts. This and the two previous networks described could all be assigned by the ISP.
  
  Note that there would be no cases of two of the networks using the same address! In the first of the smaller nets, the last octet of every address starts out with 0001. In the other smaller net, that octet would always start with 0010. In the larger net the octet always starts out as 1000.
18.16 CIDR Host Addresses
- The chart illustrates that we can assign addresses 128.211.0.17 through 128.211.0.30 to hosts in the first of the smaller networks.
- A disadvantage of the CIDR scheme is that it's not so easy to anticipate what values will show up in the dotted decimal representations of addresses.
18.17 Special IP Addresses
- A few addresses are used for special purposes. These addresses are reserved and cannot be assigned as a host address in the normal way.
- 18.17.1 Network Address
  - A network itself can be referred to by the address that begins with the network prefix and ends with all zero's.
  - For example, 18 is the network part of the class A addresses used in a certain network, and 18.0.0.0 may be used as the address of this network.
  - Another example: 130.17 is the network part of a set of class B addresses and 130.17.0.0 may be used to denote that network.
  - The network address must never be used as the destination address in a packet.
- 18.17.2 Directed Broadcast Address
  - The address that consists of "all ones" appended to the network part of an address has a special meaning too. It is an address used to send a packet to all hosts on a given network.
  - For example if the network part of the address is 130.17 then the address 130.17.255.255 is the address to use to send a packet to every host on the 130.17 network.
  - If a host outside the 130.17 network sends a packet addressed to 130.17.255.255 it will travel as a single packet across the Internet. When it arrives inside the 130.17 network, network hardware and software will see that a copy is transmitted to every host, if that network is designed to support such broadcasts.
  - Directed broadcast is sometimes misused intentionally or unintentionally, so routers and firewalls may "filter out" some or all direct broadcast packets. (For example read about smurf attacks.)
- 18.17.3 Limited Broadcast Address
  - The address consisting of all-ones (255.255.255.255) is used in some special circumstances to send a broadcast addressed to all hosts directly connected on the LAN.
  - A booting host that does not yet know its network's IP number may use a limited broadcast to ask for information.
- 18.17.4 This Computer Address
  - An IP address of all zero's (0.0.0.0) means "me."
  - A booting host will use this address as its source address when trying to discover information such as its actual IP address. (A diskless machine may have to do this, or home machine that receives its IP address from an ISP server.)
- 18.17.5 Loopback Address
  - A loopback address is a special address that a program can use to communicate with another program running on the same host.
  - When a program uses a loopback address, the packets don't get put out on the network. They just go down the protocol stack, right back up again, and then to the program assigned to the destination port number.
  - A loopback address is convenient for use when testing network software. The interaction between the client and server, or other communicating parties can be tested on one host, without requiring actual use of the network. This is expedient and it can also help isolate software-related faults from network-related problems.
  - Normal communication between different processes on a host often uses a loopback address. This has at least two advantages:
    - a program can use the same method to communicate with a program on the local host that it uses to communicate with a program on a remote host -- TCP/IP networking.
    - By using the loopback address it can avoid the overhead of actual use of the network. This saves network bandwidth and it may save in other ways such as allowing the host to use a larger packet size for the loopback address.
  - Any IP address starting with 127 is a loopback address. 127.0.0.1 is by far the most commonly used loopback address.
  - Interestingly, the loopback "network" is class A. Therefore there are about 16 million possible loopback addresses. Does that seem excessive?
18.18 Summary Of Special IP Addresses
- 0.0.0.0 -- "this computer"
- 130.17.0.0 -- network address
- 130.17.255.255 -- directed broadcast address
- 255.255.255.255 -- limited broadcast address (all ones)
- 127.0.0.1 -- loopback address
18.19 The Berkeley Broadcast Address Form
- Unfortunately there was an error in the 4.2 version of the Berkeley Standard Distribution (BSD) version of Unix.
- That software incorrectly used a network address (e.g. 130.17.0.0) for doing directed broadcast.
- The 4.2 BSD software was adopted so widely that other systems had to adapt to the incorrect use of the "Berkeley broadcast." It became a sort of "rogue standard."
- You may see a system that uses (or tolerates) "Berkely broadcast" someday, or maybe we have gotten past that.
18.20 Routers And The IP Addressing Principle
- This "technicality" is sometimes important to remember: An IP address is not really assigned to a host. An IP address is assigned to the connection between a host and the Internet.
- A router that connects two different networks will have at least two IP addresses.
- Usually the IP addresses of a router will all have different network prefixes.
- For example, at one time, on the "inside" of our campus gateway, the IP address was 130.17.1.3. On the other side, the address was 137.145.203.165.
18.21 Multi-Homed Hosts
- An ordinary host may be connected to more than one network as a convenience. The host may not do routing for other hosts. Such a host is said to be multi-homed.
18.22 Summary