HFSC queue fairness after pfSense 2.6 to 2.7 netgraph change with mixed IPv4/IPv6

Answer to the core question

In pfSense 2.7 the HFSC scheduler now creates a distinct leaf queue for IPv6 traffic under the same parent. The practical boundary that tells you whether low‑priority traffic is being starved is the leaf’s service‑curve latency compared to the parent’s 95th‑percentile latency. If the average latency of a low‑priority leaf exceeds the parent’s 95th‑percentile by more than roughly 10 ms while the WAN link is saturated, starvation is likely.

Why the 10 ms rule matters

HFSC guarantees that each leaf will not be delayed beyond the max‑delay specified in its service curve. In 2.6 IPv4 and IPv6 shared a single leaf, so the delay bound applied to both families. In 2.7 the split means each family can hit its own limit independently. When the IPv4 leaf dominates bandwidth, the IPv6 leaf can see its delay bound exceeded unless its own service‑curve parameters are tuned.

Is separate measurement of IPv4/IPv6 queue lengths required?

Yes. Because the scheduler now treats them as separate leaves, you must capture per‑leaf statistics to see which family is the bottleneck. A single aggregate queue length can hide a starving IPv6 leaf that is still short but behind a saturated IPv4 leaf.

Do service‑curve parameters need to differ for mixed traffic?

Only the IPv6 leaf typically needs a distinct configuration. The parent’s curve can remain the same, but the IPv6 leaf may require a higher max‑delay or a lower rate to keep its latency within the acceptable range. The IPv4 leaf can keep its original parameters unless you observe its own starvation.

Practical steps for your WAN link

1. Gather per‑leaf metrics – Use Diagnostics > System > Log Files > Queue Stats or tcpdump with filters to separate IPv4 and IPv6 traffic and record queue lengths and latencies while the link is under load.
2. Identify the 95th‑percentile parent latency – From the same stats, note the 95th‑percentile value for the parent HFSC queue.
3. Compare leaf latency to the parent threshold – If the low‑priority leaf’s average latency is > parent 95th‑percentile + 10 ms, starvation is likely.
4. Adjust the IPv6 leaf service curve – Reduce its rate or increase its max‑delay until the leaf’s latency falls below the threshold.
5. Re‑measure – Repeat steps 1–3 to confirm the improvement.

Missing diagnostic detail

Do you have per‑leaf queue length statistics for the saturated link?