• Internetworking
• The internet as a network of networks
• Later 2 LANs at the edges, served by routers
• Routers in core
• Transmitting
• Packetising (taking layer 4 header and data, adding layer 3 header)
• Processing and routing IP datagrams
• Fragmenting (if necessary where IP datagram > layer 2 MTU)
• Receiving
• Error checking
• Reassembly (If necessary)

• Packet-switched, connectionless
• Unreliable
• Routing
• Globally unique, delegated addresses

• IP packets routed towards destination at each router on the path
• It is connectionless, but TCP offers connection-oriented services at layer 4

• Routers forward packets on 'best effort' basis
• IP packets might get dropped, ususally due to congestion
• TCP retransmissions are inherently handled
• With UDP, application layer must handle retransmission
• Quality of service method can help
• Prioritise certain traffic

• Usually based only on destination IP address
• Routers maintain routing tables to make routing decisions

• Devices must be globally addressable to initiate IP traffic towards them
• Private address space (RFC 1918) may be used within sites

• The IP layer takes segments of data from the transport layer
• Adds IP header to create IP datagram
• TCP header and data become the data for IP datagram
• IP datagram header fields include:
• Source IP address
• Destination IP address
• IHL - header length (may vary with IP options)
• Identification field and fragment offset
• Time to live
• IP header checksum

If a link with smaller Maximum Transmission Unit (Layer 2 MTU) is encountered, the IP datagram needs to be broken down into smaller pieces

• Ethernet typically has an MTU of 1500 bytes
• In IPv4 hosts and routers can fragment IP datagrams
• IN IPv6 only sending host may fragment
• Preferrable to avoid fragmentation

• Classless Inter-Domain Routing
• Allows use of any prefix length (not just /8, /16 or /24)
• ISPs must provide proof of address requirements to RIRs
• Network address translation (NAT)
• Use private IPv4 addresses (RFC 1918) internally, global(s) externally
• Dynamic Host Configuration Protocol (DHCP) leases
• Allows IP addresses to be reused

• We want to limit propagation of layer 2 broadcast traffic
• Physical range of layer 2 media may also be limited
• Even when passed through switches
• Layer 3 (IP) routers don't forward Layer 2 Broadcasts
• Large site networks can be broken down into multiple IP subnets, each served at the edge by one router

• A site will have an IP allocation
• e.g. 152.78.0.0 - 152.78.255.255
• The allocation is a 16-bit network prefix; the remaining 16 bits can be used for host addresses and internal IP subnets
• Prefix: <first-network-address>/prefix-length
• So we would have 152.78.0.0/16
• Administrators can decide how to do the further subneting
• eg. allocating 152.78.64.0/23

• Subnet network adderss (First IP in range)
• Subnet prefix length
• Subnet mask (The fixed prefix bits for all hosts in the subnet)
• The subnet IP  broadcast address, for protocols using IP broadcasts

One address is also required for the router

• Network address (First IP in range)
• IP broadcast Address (Subnet mask)
• Router address

• Is the process of modifying IP address information in IPv4 headers while in transit across a traffic routing device.
• It is possible to run multiple hosts using private IP addresses behind a NAT device thatuses one or more global IP addresses
• Use private IP space internally
• IP headers rewritten as packets traverse NAT device
• Maps private IPs to globale IPs and vice-versa

• The original version of NAT had large pool of devices with private addresses sharing a smaller number of globally unique public addresses (Internal devices were allocated global IPs on demand)
• Today's NAT has a pool of privately addressed devices sharing one public IP

• Has only one public IP address (With many hosts wanting to share it)
• Router maintains a TCP/UDP port mapping table in addition to address mapping
• Source port rewritten in addition ot source address as packet travels out to internet
• Router device needs to maintain table of mappings
• Applications may need to use 'keepalives' where necessary to avoid the mapping state timeout

• Can share one or more global IP addresses between any number of internal hosts on network
• Often deployed for
• Perceived security of private addresses and implicit firewall
• Topology hiding, and perceived 'ease' of network renumbering
• Relies hosts behind NAT being client-only

• Poor scalability due to requiring state information (mapping between global and private IPs/ports) to be held
• Breaks end to end transparency and security (IPsec) model, because globally unique IPs are not used end to end
• Has problems if applications embed IP addresses
• Causes problems if multiple NATed sites merged, if they share common internal private IP addresses
• Adds a not insignificant cost to support NAT traversal for applications