Let’s look at an example: CESNET is the operator of Czech National Research and Education Network and also the LIR serving organisations connected to this network. CESNET obtained from RIPE NCC (RIR responsible for Europe) the prefix 2001:718::/32. /48 prefixes from this space are assigned to institutions connected to the CESNET network – for example the Czech technical University in Prague obtained the prefix 2001:718:2::/48. Its subnet number 1 has the prefix 2001:718:2:1::/64.

It is possible for the address holder to define some hierarchy of addressing scheme in the address part for which it is responsible. For example the end-site can specify the rules to assign subnet IDs. Say that the network is spread around few buildings. Natural hierarchy in this case could be to use the first part of subnet ID to identify the building and the rest to distinguish particular LANs inside the same buildings.

If you decide to deploy such a hierarchy, it is tempting to say “First N bits is the building identifier, the remaining 16–N bits identify the subnets inside building.” RFC 3531 recommends not to do so, because later you may decide that your estimation about future network development was bad and some part of the hierarchy does not suffice for the network growth.

It is better to assign upper-layer identifiers from the left side of the selected address part and growing bitwise to the right. So in our example the first building will be identified by 1000...0 bitwise (or 8000 hexadecimal), the second by 0100...0 (4000), the third by 1100...0 (c000), etc. The lower-layer identifier is then growing conventionally from right to left. So first LAN inside every building is

0...0001 (0001), the second is 0...0010 (0002), the third 0...0011 (0003), etc. When combined for example 4003 means third LAN in second building. This opposite growth ensures the postponement of part-border assignment up to the moment when the growing lengths of both parts meet. You may then find that instead of the planned division of bit 8:8 for building and LAN identifiers, the real need is to use 6:10.

Similarly the LIR can define some system in the third quadruple to assign /48 prefixes to customers. For example the first part may identify the node (point of presence), the second one distinguishes customers connected to the same node. The IPv6 addressing structure provides freedom for such decisions. The hierarchy inside individual parts marked in Figure 3-5 is not strictly defined by the RFCs, nor prohibited. It is up to the manager of the corresponding part to develop the rules. In most cases these rules are just internal, they are not communicated to the customers.

And finally we should answer the Big Principal Question: “How to get IPv6 addresses for my network?”

The answer is that it depends on your position in the addressing “food chain”. If you are an endcustomer, simply ask your IPv6 connectivity provider. The correct answer would be to ask the corresponding LIR, but the provider and LIR are usually the same body. You should get a 48-bit prefix. If the provider tries to assign you some longer prefix (smaller address space), argue and fight for /48, because RFC 3177 clearly states that /48 should be the standard network prefix length for almost all networks.

If you are a LIR and would like to provide IPv6 addresses to your customers, your situation is more complicated. You must ask your RIR to assign you some part of the address space. It is done the usual way – by filling in an application form. But there are some requirements which you have to meet to qualify for getting some address space. In the time of writing this text the requirements are:

• You have to be a LIR.
• You must not be an end site.
• You must develop a plan to provide IPv6 connectivity to connected sites. You have to assign/48 prefixes to these sites and aggregate all these prefixes to a single routing table entry used to advertise your network to the rest of world.
• You have to have a plan for assign at least 200 prefixes to other organisations within two years.

If you meet these criteria and ask for it, you should obtain a /32 prefix to manage and use for your assignments. There are running discussions about the fourth criterion, which is a little difficult to regulate. Having a plan does not mean that you really connect 200 organisations. Connecting 10 big companies means a lot more computers than connecting 300 home networks, etc. It may be the case that the fourth criterion will be abandoned someday but it is still valid for the present.