IP communication between two hosts A and B that are not on the same LAN segment must pass through one or more routers as indicated in Figure 6-1.

in Figure 6-1) travels through the sequence of routers almost unchanged. What changes from one hop to the next one are the link layer headers that encapsulate the IP datagram. The link layer headers for any two consecutive links can either be completely disparate (if the underlying link technologies are not the same) or at least differ in link layer addressing information.

We said that the datagram is delivered by the IP network from one end host to the other one almost unchanged. The routers along the forwarding path indeed do modify certain fields in the IP header but such changes are minor and well defined – for example, the hop limit in the IPv6 header (or TTL in IPv4) must be decremented at every hop. Specifically, the source and destination IP addresses in the datagram header are not changed along the way and refer to the two end hosts.

As a matter of fact, the prevailing practice of the last decade or so made the above model of transparent end-to-end communication rather rare. The current IPv4 Internet is full of various layer violating devices (firewalls, NAT gateways, proxy servers etc.) and a vast majority of connected hosts use private IPv4 addresses these days. Such an environment is hostile to certain classes of applications (peer-to-peer, IP telephony etc.) and has also considerable drawbacks from the viewpoint of network reliability and management. As discussed in [RFC2775], several factors are behind this significant architectural shift. One of them has been the lack of global IPv4 addresses and this is where IPv6 could really help. However, layer violating devices are often deployed for other purposes, most notably security, and so devices like firewalls are likely to persist in the IPv6 Internet as well (see Chapter 9, Security).