Providing 64 bits to identify the interface in the scope of a single subnet seems to be a huge extravagance. For example 48 bits are sufficient for Ethernet addresses which are world-wide unique. Subnets for which 16 bits would not suffice to identify all the nodes are hard to imagine. On the other hand this 64-bit long interface identifier simplifies significantly some autoconfiguration mechanisms.

RFC 3513 specifies the use of modified EUI-64 identifiers in this part of the IPv6 address. EUI-64 is a network interface identifier defined by the IEEE. The modification deployed in IPv6 is related to 7th bit of the 64-bit identifier. This bit distinguishes global identifiers (world-wide unique) from the local ones (unique only in the scope of single link). The value of this bit is inverted in IPv6 addresses. Hence the value 0 of this bit means a local identifier, while a value of 1 indicates a global ID. How do we determine the value of this final part of the IPv6 address? The answer depends on the lower-layer address which the corresponding interface has. Basic rules are following:

Interface has an EUI-64 identifier

This is the simplest case. The 7th bit of the existing EUI-64 identifier is inverted and the resulting value is used.

Interface has a MAC (Ethernet) address

There is a simple algorithm converting the MAC address into a modified EUI-64: the global flag (7th bit) of the MAC address is inverted and the value fffe is inserted between the 3rd and 4th byte of the MAC address. For example the MAC address 00:8c:a0:c2:71:35 is converted to interface ID 028c:a0ff:fec2:7135 (the conversion is illustrated in Figure 3.3).

Otherwise

In other cases the network manager simply assigns some identifier to the interface. Typically some simple identifiers (like 0:0:0:1 and 0:0:0:2) are used. Such artificial identifiers are used for example for serial lines, which do not provide any values usable as a ground for the
identification.

From a technical point of view this is a perfect working mechanism. But there is a hidden drawback - a threat to privacy. Since a common computer is equipped by some MAC-addressed network card, the second rule is used for the vast majority of computers. But this means that even if the user is travelling and changing the networks used to connect to the Internet, the interface identifier of his/her computer remains constant. In other words the computer can be tracked.

RFC 3041 [RFC3041] solves this problem. It recommends the interface to have several identifiers. One of them is a fixed, EUI-64 based identifier. This is used in the “official” (DNS registered) address and is used mainly for incoming connections. The additional identifiers are randomly generated and their lifetime can be limited to a few hours or days. These identifiers are used for outgoing connections, initiated by the computer itself. Thanks to these short-lived identifiers the systematic long-term tracking of computer activities is much more difficult.