Address Configuration and Translation: DHCP and NAT

Dynamic Host Configuration Protocol (DHCP)

A host's IP address can be configured manually by a system administrator, but it is more commonly obtained automatically with the Dynamic Host Configuration Protocol (DHCP). DHCP makes joining a network "plug and play." Besides an IP address, it tells the host its subnet mask, the address of its default gateway (first-hop router), and the address of its local DNS server. A DHCP address comes with a lease time; it must be renewed before the lease expires or it is reclaimed for other hosts.

DHCP is a client-server protocol that runs over UDP: clients use port 68 and servers use port 67. A new host obtains an address in four steps:

At first the new host has no address of its own and does not know any server's address, so the discover message has source address 0.0.0.0 and the limited-broadcast destination 255.255.255.255. There may be several DHCP servers on the LAN, and each may reply with an offer. The client echoes the offer it chose in its request, which it still broadcasts so that every server learns the choice and unchosen servers can reclaim the addresses they held in reserve.

When half the lease has passed ($T_1 = 0.5 \times$ lease), the client unicasts a renewal request directly to its server. If no answer arrives by $T_2 = 0.875 \times$ lease, it broadcasts the request so that any server can extend the lease. If the lease expires with no answer, the client must stop using the address and start over with a discover message.

Network Address Translation (NAT)

Section 4.3 showed that the private ranges 10.0.0.0/8, 172.16.0.0/12, and 192.168.0.0/16 are used only inside local networks and cannot appear on the public Internet. Yet hosts in homes and enterprises clearly need Internet access. Network Address Translation (NAT) solves this. A NAT router has a single public address on its Internet-facing side, while the inside network uses private addresses. To the outside world, the whole inside network looks like a single device with a single IP address.

The NAT router maintains a NAT translation table. Each entry associates one public port number with one inside (IP address, port number) pair:

1. Inside host 10.0.0.1 sends a datagram from source port 3345 to the web server 128.119.40.186:80.
2. The NAT router picks an unused port number, 5001, rewrites the source address and port to 138.76.29.7:5001, and records 5001 ↔ 10.0.0.1:3345 in its translation table.
3. The web server sees only 138.76.29.7:5001, so its reply is addressed to 138.76.29.7:5001.
4. When the reply reaches the NAT router, the router looks up destination port 5001 in the table, rewrites the destination address and port back to 10.0.0.1:3345, and forwards it to the inside host.

The port-number field is 16 bits long, so a single public address can support more than 60,000 simultaneous connections.

NAT has costs. Translation entries are created only when inside hosts open connections outward, so outside hosts cannot initiate connections to servers inside unless a static entry such as port forwarding is configured on the NAT. Port numbers are meant to identify processes, not hosts, and a router rewriting them violates the layering principle that the network layer should process only up to the network-layer header. It also complicates applications such as P2P that need direct connections, which require NAT traversal techniques. In the long run, the address shortage should be solved by IPv6 (see Section 4.7).