Before explaining the network layer in detail, it is useful to begin by analyzing the service provided by the `datalink` layer. There are many variants of the datalink layer. Some provide a connection-oriented service while others provide a connectionless service. In this section, we focus on connectionless datalink layer services as they are the most widely used. Using a connection-oriented datalink layer causes some problems that are beyond the scope of this chapter. See :rfc:`3819` for a discussion on this topic.
There are three main types of datalink layers. The simplest datalink layer is when there are only two communicating systems that are directly connected through the physical layer. Such a datalink layer is used when there is a point-to-point link between the two communicating systems. The two systems can be hosts or routers. :abbr:`PPP (Point-to-Point Protocol)`, defined in :rfc:`1661`, is an example of such a point-to-point datalink layer. Datalink layers exchange `frames` and a datalink :term:`frame` sent by a datalink layer entity on the left is transmitted through the physical layer, so that it can reach the datalink layer entity on the right. Point-to-point datalink layers can either provide an unreliable service (frames can be corrupted or lost) or a reliable service (in this case, the datalink layer includes retransmission mechanisms similar to the ones used in the transport layer). The unreliable service is frequently used above physical layers (e.g. optical fiber, twisted pairs) having a low bit error ratio while reliability mechanisms are often used in wireless networks to recover locally from transmission errors.
The second type of datalink layer is the one used in Local Area Networks (LAN). Conceptually, a LAN is a set of communicating devices such that any two devices can directly exchange frames through the datalink layer. Both hosts and routers can be connected to a LAN. Some LANs only connect a few devices, but there are LANs that can connect hundreds or even thousands of devices.
In the next chapter, we describe the organization and the operation of Local Area Networks. An important difference between the point-to-point datalink layers and the datalink layers used in LANs is that in a LAN, each communicating device is identified by a unique `datalink layer address`. This address is usually embedded in the hardware of the device and different types of LANs use different types of datalink layer addresses. Most LANs use 48-bits long addresses that are usually called `MAC` addresses. A communicating device attached to a LAN can send a datalink frame to any other communicating device that is attached to the same LAN. Most LANs also support special broadcast and multicast datalink layer addresses. A frame sent to the broadcast address of the LAN is delivered to all communicating devices that are attached to the LAN. The multicast addresses are used to identify groups of communicating devices. When a frame is sent towards a multicast datalink layer address, it is delivered by the LAN to all communicating devices that belong to the corresponding group.
The third type of datalink layers are used in Non-Broadcast Multi-Access (NBMA) networks. These networks are used to interconnect devices like a LAN. All devices attached to an NBMA network are identified by a unique datalink layer address. However, and this is the main difference between an NBMA network and a traditional LAN, the NBMA service only supports unicast. The datalink layer service provided by an NBMA network supports neither broadcast nor multicast.
Unfortunately no datalink layer is able to send frames of unlimited side. Each datalink layer is characterized by a maximum frame size. There are more than a dozen different datalink layers and unfortunately most of them use a different maximum frame size. The network layer must cope with the heterogeneity of the datalink layer.
In the late 1980s and early 1990s the growth of the Internet was causing several operational problems on routers. Many of these routers had a single CPU and up to 1 MByte of RAM to store their operating system, packet buffers and routing tables. Given the rate of allocation of IPv4 prefixes to companies and universities willing to join the Internet, the routing tables where growing very quickly and some feared that all IPv4 prefixes would quickly be allocated. In 1987, a study cited in :rfc:`1752`, estimated that there would be 100,000 networks in the near future. In August 1990, estimates indicated that the class B space would be exhausted by March 1994. Two types of solution were developed to solve this problem. The first short term solution was the introduction of Classless Inter Domain Routing (:term:`CIDR`). A second short term solution was the Network Address Translation (:term:`NAT`) mechanism, defined in :rfc:`1631`. NAT allowed multiple hosts to share a single public IPv4 address.
However, in parallel with these short-term solutions, which have allowed the IPv4 Internet to continue to be usable until now, the Internet Engineering Task Force started working on developing a replacement for IPv4. This work started with an open call for proposals, outlined in :rfc:`1550`. Several groups responded to this call with proposals for a next generation Internet Protocol (IPng) :
The IETF decided to pursue the development of IPng based on the SIPP proposal. As IP version `5` was already used by the experimental ST-2 protocol defined in :rfc:`1819`, the successor of IP version 4 is IP version 6. The initial IP version 6 defined in :rfc:`1752` was designed based on the following assumptions :
When the work on IPng started, it was clear that 32 bits was too small to encode an IPng address and all proposals used longer addresses. However, there were many discussions about the most suitable address length. A first approach, proposed by SIPP in :rfc:`1710`, was to use 64 bit addresses. A 64 bits address space was 4 billion times larger than the IPv4 address space and, furthermore, from an implementation perspective, 64 bit CPUs were being considered and 64 bit addresses would naturally fit inside their registers. Another approach was to use an existing address format. This was the TUBA proposal (:rfc:`1347`) that reuses the ISO CLNP 20 bytes addresses. The 20 bytes addresses provided room for growth, but using ISO CLNP was not favored by the IETF partially due to political reasons, despite the fact that mature CLNP implementations were already available. 128 bits appeared to be a reasonable compromise at that time.
