IP

• IP header
• packet forwarding
• routing function
• IP addressing
• local configuration

DNS Requirements

• send a packet to an arbitrary host in the network -- packets must be forwarded to destination identified by address
• packet size could be large if we batch requests or names are long -- fragmentation
• data loss, corruption, or reordering are not fatal because DNS itself can retransmit
• these requirements define rather well the Internet Protocol

IP header

    0                   1                   2                   3   
    0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |Version| HLen  |Type of Service|          Total Length         |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |         Identification        |Flags|      Fragment Offset    |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |  Time to Live |    Protocol   |         Header Checksum       |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                       Source Address                          |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                    Destination Address                        |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                    Options                    |    Padding    |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

IP header fields

• all fields are big-endian
• version number must be first field (so IP can evolve)
• Type of Service is defined but not widely used
• Total Length is needed for protocols that may pad the packet (e.g. ethernet) and to check for errors
• header length used for IP options, e.g. timestamp
• TTL/Time To Live (really a "maximum hop count") kills packet in case of routing errors
• protocols: 1 (ICMP), 6 (TCP), 17 (UDP)

More IP header fields

• header checksum (RFC 1071, http://www.ietf.org/rfc/rfc1071.txt) only protects header
• source address might be useful for responding
• packet ID and fragmentation fields (discussed next lecture) allow us to send packets larger than the underlying network's MTU (Maximum Transmission Unit)
• minimum MTU for carrying IP is 576 bytes

Some properties of IP

• connectionless: routing is based only on the destination address
• TTL field keeps small routing mistakes from bringing down the network
• unreliable: no packet sequence numbers, acknowledgements, error checking on the data
• source address is never used in "normal" processing

IP forwarding

• In this figure, if A sends to E, A puts E in the destination address field, and sends to B
• B chooses an interface (3) and sends to F
• F chooses an interface (1) and sends to H
• H chooses an interface (1) and sends to A
• if H chooses the wrong interface, the packet will never be delivered

Definitions

• a host with multiple interfaces is multihomed
• a host with multiple interfaces that agrees to forward packets is a router (formerly a gateway)

IP Router Algorithm

1. build and maintain a table of pairs: (destination address, interface)
2. when a packet is received with destination A, check the checksum. If the checksum is incorrect, discard the packet
3. if A is the address of one of my interfaces, process the packet
4. otherwise, decrement the TTL (discard if new TTL=0)
5. search for A in my routing table
6. if found (A, I), forward the packet on interface I, otherwise, discard the packet

Routing Errors

• Routing table has more than one entry for a single destination (this is generally OK)
• A destination might be connected, but not be in the table -- no communication is possible
• A packet is routed in the wrong direction, but eventually gets there (not uncommon, OK)
• A packet is routed in the wrong direction, and either starts to loop or ends up at the wrong place, so the packet is lost -- no communication, packet is discarded when TTL reaches zero
• all errors (except physical disconnection) are in the routing table

Address Groupings

• If the routing table needed an entry for every host, it might be very large (up to 2³² entries)
• so we group entries by the initial (high-order) bits, representing the network number
• only one routing table entry is needed per network
• the remaining bits (the host number) are used, in a network-dependent way, only within a network, i.e. are of no concern to IP
• this summarization is most effective when the network structure is hierarchical

IP addresses

• two ways of saying which part is the network and which part the host number:
  • class-based: the first few bits tell us how many bits are in the network part (class A: 8 bits, class B: 16 bits, class C: 24 bits). This is the older way of doing this (but is the standard way in IPv6).
  • class-less (newer): each routing table entry also has a mask, a 32-bit number of the form 111...1100...00 that has:
    • a 1 bit for every bit of the address that is part of the network number, and
    • a 0 bit for every bit of the address that is part of the host number
    sometimes we use a number (0..30) instead of a 32-bit mask, e.g. 128.171.10.1/255.255.255.0 can be written as 128.171.10.1/24
  this is CIDR, Classless Inter-Domain Routing

Classless Inter-Domain Routing

• CIDR is a more efficient way of using IP addresses, because you can:
  • have network sizes other than 2⁸, 2¹⁶, and 2²⁴ addresses
  • do multiple hierarchical subdivisions, e.g. 128.171.0.0/16 for routing to UH, and 128.171.10.0/24 for routing within UH
• CIDR was adopted around 1994, due to impending exhaustion of class B addresses

Local Configuration

• each interface must be given its IP address
• host/router must place all the local routes (and network masks for CIDR) into the routing table
• host/router must know the address of the "default" router
• there may be further configuration for DNS, particularly the IP number(s) of DNS servers

Summary

• the Internet Protocol is designed to take data end-to-end under a "best effort" model
• IP does not provide:
  • reliability
  • in-order delivery
  • error-free delivery
• routing is easy once the tables are built
• summarizing (routing to networks instead of hosts) helps reduce the size of the tables