Testing and validating IPv6 readiness
Write an IPv6 readiness test plan covering network, services and applications. Measure and compare IPv6 and IPv4 performance. Explain how Happy Eyeballs (RFC 8305) can hide IPv6 faults and how address selection (RFC 6724) affects results.
Take the quiz · pass mark 70% · 9 questions
Explanation
The last step of any transition is proving it works — and proving it does not merely appear to work because clients quietly fall back to IPv4.
A readiness test plan
Test in three layers:
- Network: every segment has correct RAs/addresses, routing both ways, PMTUD working (no MTU black-holes from L6.02/L6.06), DNS returning AAAA, and firewall policy permitting the right ICMPv6 (L5.02). Verify reachability to native IPv6 and, where used, through NAT64/CLAT.
- Services: each service listens on IPv6 (not just IPv4), has an AAAA record, presents correct certificates for the IPv6 endpoint, and logs the real client address canonically (L6.07).
- Applications: the app opens IPv6 sockets, handles literals and zone IDs, and works in an IPv6-only environment, not just dual stack.
Happy Eyeballs hides faults (RFC 8305)
Happy Eyeballs v2 (RFC 8305) makes dual-stack clients fast and robust: the client resolves A and AAAA, and races connections, preferring IPv6 but falling back to IPv4 quickly if IPv6 is slow or fails. That is great for users and terrible for testing, because a broken IPv6 path is invisible — the user still succeeds over IPv4. Key timers:
- Resolution Delay (default 50 ms): after the first address family resolves, wait briefly for the AAAA before starting, so IPv6 gets a fair chance.
- Connection Attempt Delay (default 250 ms, min 100 ms, max 2 s): how long to wait for the preferred (IPv6) connection before trying the next candidate (IPv4).
So a site can look "IPv6-enabled" while every real connection silently uses IPv4. To test IPv6 honestly you must disable fallback or test IPv6-only: use an IPv6-only client/network, force the address family (curl -6), or measure per-family explicitly.
Address selection (RFC 6724)
RFC 6724 governs which source and destination address a host picks when several are available. It generally prefers IPv6 over IPv4 and native over transition addresses, and prefers temporary over stable for privacy (L5.06). Surprising test results — IPv4 used when you expected IPv6, or the "wrong" source address — are often RFC 6724 policy, so check the default address selection and any local policy table before blaming the network.
Measuring performance
Compare IPv6 and IPv4 separately, not whatever Happy Eyeballs chose: run the same request over each family and compare latency, throughput and success rate. A blackbox exporter (e.g. Prometheus) probing both families, or curl -4/curl -6 in a loop, exposes an IPv6 path that is reachable but slower — the kind of fault Happy Eyeballs masks.
ictNetworks scenario
ictNetworks' readiness plan tests network, services and apps per segment, and crucially runs an IPv6-only probe (and curl -6) so Happy Eyeballs cannot hide a broken AAAA or a NAT64 fault. Its Prometheus blackbox exporter probes each service over IPv4 and IPv6 separately and alerts on per-family divergence, and it checks RFC 6724 selection when a host unexpectedly prefers IPv4. A readiness sign-off requires the IPv6-only test to pass, not just the dual-stack one.
Drawing diagram…
Worked examples
Example 1: Resolution Delay
After the A record arrives, Happy Eyeballs v2 waits a Resolution Delay of 50 ms (default) for the AAAA before starting, so the client does not rush into IPv4 when IPv6 is a moment behind in DNS (RFC 8305).
Example 2: Connection Attempt Delay
Having started the IPv6 connection, the client waits a Connection Attempt Delay of 250 ms (default) before also trying IPv4. A broken IPv6 path therefore costs the user only ~250 ms — and hides the fault from casual testing.
Example 3: The floor you must still meet
Even when everything else looks fine, the IPv6 path must carry the minimum link MTU of 1280 bytes end to end (RFC 8200's floor). A readiness test that skips PMTUD can pass while large transfers black-hole, so include a large-packet test.
Lab
Goal: Run the readiness plan against the lab and find a fault that Happy Eyeballs was masking.
You need: the containerlab reference topology with a deliberately introduced IPv6 fault (e.g. a missing AAAA, an MTU black-hole, or a NAT64 gap), a dual-stack client, an IPv6-only client, curl, and a Prometheus blackbox exporter (or a scripted per-family probe).
- From the dual-stack client, access the services and observe everything "works" (Happy Eyeballs fallback).
- Switch to the IPv6-only client (or
curl -6/ disable fallback) and find the service(s) that actually fail over IPv6. - Use the per-family blackbox probe to show the IPv6 vs IPv4 divergence (latency/success) that the dual-stack test hid.
- Diagnose the root cause (DNS AAAA, PMTUD/MTU, ICMPv6 filter, or NAT64) and fix it.
- Re-run the IPv6-only test and the per-family probe to confirm parity; record the before/after.
What to submit: the test plan, the dual-stack-vs-IPv6-only evidence, the per-family probe output, the root-cause diagnosis and fix, and the confirmation of parity.
Summary
- Test network, services and applications, and always include an IPv6-only test, not just dual stack.
- Happy Eyeballs v2 (RFC 8305, Resolution Delay 50 ms, Connection Attempt Delay 250 ms) hides IPv6 faults behind fast IPv4 fallback.
- Force the family (
curl -6, IPv6-only client) and probe each family separately to see the truth. - RFC 6724 address selection explains "unexpected" family/source choices; the IPv6 path must still carry 1280 bytes.
- This completes Level 6; next is Level 7 (deployment in ISP, enterprise, data-centre and cloud networks).
Quiz imported · 9 questions · pass mark 70%. Lab questions are not auto-marked.