Tune OPNsense FQ_CoDel bandwidth against latency under load

Bottom line: Choose the shaping rate from an agreed tradeoff between throughput and latency during load. Keep download and upload observations separate, and preserve the unshaped baseline so the owner can assess what changed.

Source fact: what the authority says

The OPNsense FQ_CoDel guide describes separate download and upload pipes, queues and rules. Its tuning procedure begins with several unshaped speed tests to record average rates and latency, then uses an initial bandwidth setting of 85 percent of the advertised rate. The procedure raises a directional bandwidth setting while latency remains low and backs it off when latency increases, then repeats for the other direction.

What the source does not establish

The example and starting percentage do not establish suitable values for a particular circuit. This article concerns the documented bandwidth-tuning decision, not a universal queue profile. DSE leaves other advanced parameters, service-level claims and application acceptance subject to their own evidence.

Applicability questions

• Which link and direction form the intended bottleneck, including any separate upstream equipment?
• What repeatable test and observation interval will represent the operational concern?
• What throughput and loaded-latency tradeoff will the service owner accept?
• Who can authorize test load, observe user impact and restore the prior shaping settings?

DSE recommendation: controlled next steps

The following steps are DSE recommendations based on the cited source.

DSE recommends an initial worksheet with separate upload and download columns. Record the circuit description, current shaping configuration, test endpoint, time and ordinary workload conditions. Run an approved unshaped baseline before choosing a proposed setting. Keep all baseline observations, including an inconvenient slow or variable result, instead of selecting only the best-looking sample.

Use an attended trial with one controlled bandwidth adjustment at a time. For a proposed example, compare a baseline, a conservative shaping setting and a cautiously higher setting using the same method. Have the reviewer decide which latency change warrants backing off. Treat the percentage in the guide as a documented starting procedure rather than a substitute for that local decision.

Keep the two directions distinct through the review and retain the actual final values. Avoid changing unrelated queue parameters during the same trial unless the expanded scope has its own justification. After the directional tuning, ask the application owner to assess representative calls, interactive work or transfers under the approved observation plan. Do not infer an application result solely from a speed-test grade.

Keep simultaneous and single-direction load trials separately labeled if both are approved. State which trial drives the acceptance decision and why. Review any busy-period observation on its own terms rather than blending all readings into one favorable average.

Verification and evidence

Proposed verification: repeat the chosen test at each recorded setting and compare the resulting rate and latency observations with the unshaped baseline. Annotate competing workload, any interruption and any method change. Use the same test service throughout the comparison unless a separate check explains why another endpoint is needed.

Retain the sequence of settings and observations so a second administrator can identify the selected tradeoff. Include a return to the previous setting if the pilot becomes unacceptable, then verify the restored configuration. Record the application owner’s observed acceptance separately. Reopen the tuning decision after a circuit-rate or traffic-path change rather than assuming the earlier values remain suitable.

Official references

• OPNsense — Fighting Bufferbloat with FQ_CoDel — official product documentation; confirm the installed release and configuration.

Source review completed October 28, 2026. Recheck the official documentation and deployment-specific applicability before a production change.