What is bufferbloat?

Why capacity and responsiveness can disagree

A busy link may queue packets faster than it can send them. Large or poorly managed queues can make interactive traffic wait behind bulk transfers. Upload is worth checking separately: a backup can fill a relatively small upstream link while download capacity remains available.

A browser observes more than the queue at your router. Wireless contention, device processing and the remote endpoint can affect request timings too. An increase under load is therefore a symptom consistent with bufferbloat, not a direct queue measurement.

Understand this tool’s grade

The calculation takes the larger of the download-loaded and upload-loaded median latency, subtracts idle median latency, and clips a negative result to zero. These are the current local grading rules; they are not an industry certification or another website’s score.

An A+ does not guarantee a fault-free connection. It describes a short comparison to this endpoint. With idle 18 ms, download-loaded 60 ms and upload-loaded 140 ms, the increase used is 122 ms and the grade is C.

A useful before-and-after experiment

1. Use a capable wired device if possible. Record the server, duration, device and time.
2. Take repeat baseline results with nonessential traffic paused. Note both loaded directions, not just the letter.
3. Reproduce the actual workload and check whether the application also suffers. A synthetic multi-request transfer is not identical to every real workload.
4. Try one reversible change: schedule a backup, set an application upload limit, or adjust a supported router feature using its documentation.
5. Repeat under comparable conditions and compare delay, throughput and application quality. Keep the earlier settings so you can undo a change.

What queue management can change

Smart Queue Management, where supported, combines traffic shaping and queue-management mechanisms to control where congestion occurs and how traffic shares the link. A router’s generic QoS label does not establish that it provides the same behaviour. See OpenWrt’s SQM overview for one documented implementation.

Shaping below a stable bottleneck rate may help the router manage the queue. For an illustrative stable 100/20 Mbps connection, a 90% trial would be 90 Mbps down and 18 Mbps up. That is a starting experiment, not a universal optimal setting. Variable-rate access, link overheads and router CPU limits can require a different approach. Use the documentation for your supported equipment; this guide does not require installing replacement firmware.

Choose a response from the evidence

• If limiting one backup restores a call, scheduling or that application’s rate control may be sufficient.
• If poor responsiveness repeatedly follows a saturated link, assess supported queue management and the throughput trade-off.
• If wired idle performance is also unstable, investigate that separately rather than assuming queue management will cure it.
• If only one room or wireless client struggles, compare coverage and radio conditions before changing the broadband package.

A faster service may provide more spare capacity for a particular workload, but it does not automatically repair queue management, interference or faults. Keep evidence of the actual symptom alongside the benchmark.