Addressing hierarchy, summarisation and IPv6 migration
What you will learn in this module. You will learn to design the routing layer of an enterprise network the way the CCNP ENSLD 300-420 exam expects: start from the business requirements, list the constraints, compare the candidate options, and justify one design. This module covers structured IPv4 and IPv6 addressing, OSPF, EIGRP, IS-IS and BGP design, redistribution and filtering, fast convergence, and IPv6 migration options. Each chapter begins with a customer case. Prerequisites: ENCOR-level knowledge of subnetting, OSPF, EIGRP and BGP basics. We do not re-teach configuration from zero; we ask why this design and not another.
Customer case: Harbourline Logistics
Harbourline Logistics has 6,500 staff, three campuses (Northbridge, Eastport, Westfield), 42 branch depots and two data centers. Last year it bought a smaller company, Tidewater Freight, which also uses 10.0.0.0/8. Harbourline won a government contract that requires IPv6 within 18 months. The network team has one more problem: branch routers have limited memory and no budget for replacement this year.
Requirements: every region must be summarised so that the WAN routing table stays small; there must be room for 5 years of growth; IPv6 must follow the same logic as IPv4; Tidewater must be merged without a long outage. Constraints: branch router memory, 18-month IPv6 deadline, overlapping address space, a small team.
Why hierarchy comes first
Routing protocols cannot summarise what the address plan did not group. A hierarchical addressing plan gives each region one contiguous block, each site a sub-block of it, and each VLAN a subnet of that. The region router then advertises one summary route instead of hundreds of subnets. This gives three design benefits the exam tests: scalability (smaller tables, smaller link-state databases), stability (a flapping subnet inside a site is hidden behind the summary, so the failure domain shrinks and the blast radius of a change is small) and faster convergence (fewer routes to recompute).
One block per region and per site: the region router advertises a single summary to the core.
Candidate options
| Option | How it works | Strength | Weakness |
|---|---|---|---|
| A. Allocate on request | Next free subnet to whoever asks | Zero planning effort | No summarisation; table grows with every site |
| B. Geographic hierarchy | Region block, then site block, then function subnet | Summaries at region and site boundaries | Needs up-front planning and reserved growth space |
| C. Function-based blocks | One block for all voice, one for all Wi-Fi, one for all servers | Easy to write ACLs and QoS policy | A site is spread over many blocks; summarisation by location is impossible |
Decision: option B, with a small amount of function logic inside each site block (the third octet, or the /24 index, identifies data, voice, Wi-Fi, guest, management). Policy by function is then done with ACLs on a regular pattern, while routing follows geography.
The chosen IPv4 plan
Worked example. From 10.0.0.0/8: infrastructure 10.0.0.0/14 (loopbacks as /32 from 10.0.0.0/20, point-to-point links as /31 from 10.0.16.0/20), Region North 10.4.0.0/14, Region East 10.8.0.0/14, Region West 10.12.0.0/14, data centers 10.16.0.0/14 and renumbered Tidewater 10.20.0.0/14. A /14 split into /20 sites gives 2^(20-14) = 64 sites per region, and each /20 holds 16 subnets of /24. Harbourline needs 14 depots per region today, so the plan keeps more than 4 times spare capacity. Every boundary falls on a bit boundary, so each region and each site is one summary.
Point-to-point links use /31 (RFC 3021), saving half the addresses; loopbacks are /32. Leave unused blocks unallocated and written down: reserve, do not fill.
The IPv6 plan follows the same logic
IPv6 removes the shortage, so the goal is a clean hierarchy, not saving addresses. Assume the provider gives Harbourline 2001:db8::/32 (a documentation prefix, used here as the example). Give each region a /40 (256 regions possible), each site a /48 (256 sites per region) and each VLAN a /64 (65,536 per site). Align on nibble boundaries so summaries are readable: Region North is 2001:db8:100::/40, and site 5 in North is 2001:db8:105::/48, matching IPv4 site 10.4.80.0/20. Always use /64 on host subnets (SLAAC needs it); use /127 on point-to-point links (RFC 6164) and /128 for loopbacks.
! Dual-stack core link: /31 and /127 ipv6 unicast-routing interface TenGigabitEthernet0/0/1 description P2P to Core1 ip address 10.0.16.0 255.255.255.254 ipv6 address 2001:DB8:0:1::/127 ! Region router: advertise one IPv4 summary toward the core (OSPF area range) router ospf 1 area 10 range 10.4.0.0 255.252.0.0
Core1# show ip route 10.4.0.0 255.252.0.0 O IA 10.4.0.0/14 [110/21] via 10.0.16.0, 00:04:12, TenGigabitEthernet0/0/1 RegionN# show ip route 10.4.0.0 255.252.0.0 O 10.4.0.0/14 is a summary, 00:04:12, Null0
IPv6 migration options
The blueprint groups migration into three families. Native (dual stack): run IPv4 and IPv6 side by side on every device; simplest to operate and the preferred end state, but it doubles routing tables, security policy and monitoring. Overlay (tunnelling): carry IPv6 across an IPv4-only core using manual tunnels, GRE, or ISATAP-style methods; useful for a quick start but adds MTU and troubleshooting overhead. Boundary (translation): NAT64 with DNS64 lets IPv6-only clients reach IPv4-only servers; it suits an IPv6-only edge or data center. Harbourline chooses dual stack in campuses and branches (the hardware supports it), and a short-lived tunnel only for one branch whose provider cannot yet carry IPv6.
| Method | Use when | Main risk |
|---|---|---|
| Dual stack | Devices and carriers support both | Twice the policy and tables |
| Tunnelling | Core or provider is IPv4 only | MTU, extra hops, hard to debug |
| NAT64 + DNS64 | IPv6-only clients need IPv4 content | Breaks apps with embedded IPv4 literals |
Post-mortem: the plan that was never written
Before this project, Harbourline allocated /24s on request. After eight years the WAN routing table held 1,900 routes. When the Eastport depot link flapped, every branch router re-ran SPF and two small routers ran out of memory and reloaded. Because nothing could be summarised, the failure domain was the whole company. The hierarchical plan reduced each branch's table to about 12 routes (regional summaries plus a default) and the same flap now stays inside Eastport.
Lesson: address planning is a routing design decision. You cannot fix a missing hierarchy later with configuration alone.
Common mistakes. Summarising a block that has holes: traffic for the missing part is attracted to the summary and dropped (a black hole). Using /64 not at all on LANs, or /126 where /127 is intended. Planning IPv6 as a separate project with a different structure from IPv4, which doubles the operational effort.
Exam trap. Questions describe a flat or random plan and ask what limits scalability: the answer is the lack of contiguous, bit-aligned blocks that can be summarised. Summarisation reduces table size, query or SPF scope and the blast radius of a failure. Remember the three migration families: native, overlay, boundary.
"How would you design the addressing for a company with 40 branches that must add IPv6 next year?"
Strong answer: give each region a contiguous block, each site a sub-block, each VLAN a subnet, aligned on bit boundaries so each level is a single summary; reserve growth; use /31 or /127 on links and /32 or /128 on loopbacks; mirror the same hierarchy in IPv6 with /40 per region, /48 per site, /64 per VLAN; choose dual stack and use tunnels or NAT64 only where a boundary requires it.
Key takeaways
- Design addressing before routing: summarisation needs contiguous, bit-aligned blocks.
- Hierarchy gives scalability, stability and smaller failure domains.
- Use geographic blocks for routing and a regular pattern inside a site for policy.
- IPv6 plan: /48 per site, /64 per subnet, /127 on links, aligned on nibbles.
- Migration families: native dual stack, overlay tunnels, boundary translation (NAT64).