becarpenter-book6/5. Network Design/5. Network Design.md

# Network Design

A first very general remark is that since IPv6 is a datagram protocol,
whose routing relies on longest matching of address prefixes, the
highest level of design decisions are identical to those for IPv4.

There is one constraint that does not apply to IPv6: there is
effectively no theoretical limit to the number of hosts per subnet.
(Mathematically, there is a limit of about 18.10<sup>18</sup> nodes on a
/64 subnet, but this is of no practical concern.) However, most network
designers will never place hundreds or thousands of hosts on a single
subnet, for performance reasons.

A network designer does, however, have more flexibility with IPv6. If an
enterprise has a /48 prefix, 16 bits are available to identify more than
65 thousand individual subnets, a luxury for most IPv4 network
designers.

Setting these details aside, there is no reason why an IPv6 network will
have a different macroscopic design than an IPv4 network. The detailed
approach will vary.

- If the intention is a "retrofit" where IPv6 support is added to an
  existing IPv4 network, major topology will not change, but items such
  as border routers, firewalls, interior routers, and DMZs will need to
  be upgraded accordingly. Clearly, a specific choice of IPv6/IPv4
  coexistence mechanism must be made, and applied consistently. In the
  past, most networks have chosen the original dual-stack approach
  \[[3. Dual stack scenarios](../3.%20Coexistence%20with%20Legacy%20IPv4/Dual%20stack%20scenarios.md)\]
  but designers should now also consider
  [3. Translation and IPv4 as a service](../3.%20Coexistence%20with%20Legacy%20IPv4/Translation%20and%20IPv4%20as%20a%20service.md).
  A priority will be adding comprehensive IPv6 support to the NOC and
  all its systems, before deployment to users. An equal priority will be
  training of all NOC and support personnel: they need to be IPv6
  evangelists.

- If the intention is a "greenfield" deployment with no existing IPv4
  network, the main topology will be conventional, but a specific choice
  of mechanism for IPv4 as a service must be made
  \[[3. Translation and IPv4 as a service](../3.%20Coexistence%20with%20Legacy%20IPv4/Translation%20and%20IPv4%20as%20a%20service.md)\].
  The NOC must be designed from the start based on IPv6, with the
  ability to manage IPv4 as a service.

- A specific difference between a retrofit design and a greenfield design
  is that an existing IPv4 network almost inevitably has subnets limited
  to 256 or fewer hosts, often as few as 64. Since the normal subnet
  prefix in IPv6 is a /64, there is no such limitation in a greenfield
  deployment of IPv6. However, for
  practical reasons such as the rate of link-local multicasts, very large
  subnets should be avoided. As noted elswehere
  \[[2. Address resolution](../2.%20IPv6%20Basic%20Technology/Address%20resolution.md)\]
  this applies particularly to wireless networks.

This chapter continues with a discussion of address planning, inevitably
combined with subnet design.

## [Address Planning](Address%20Planning.md)

<!-- ## Name (add plain section names like that) -->

<!-- Link lines generated automatically; do not delete -->

### [<ins>Back to main Contents</ins>](../Contents.md)