Quick answer: how to fix bufferbloat
Bufferbloat is mitigated by configuring Smart Queue Management (SQM): such as CAKE, a shaping-capable queue discipline that combines flow isolation, active queue management and traffic prioritization, or FQ-CoDel (RFC 8290). SQM can reduce delay at a bottleneck it effectively controls, depending on conditions and setup. By shaping throughput slightly below sustainable line capacity (typically 85% to 90%, per OpenWrt guidance), it aims to keep the managed bottleneck in the router rather than in an unmanaged modem or ISP buffer. Fair queueing isolates competing data flows to prevent single bulk transfers from dominating the connection, while active queue management keeps standing queues short. Explicit traffic prioritization depends on the chosen queue discipline, packet classification, markings, and configuration rather than automatic identification. Before modifying router settings, verify whether queueing delay originates on your local router, your modem/ONT, or a congested Wi-Fi connection.
If voice calls, video meetings, or online games degrade while someone downloads a large file or streams 4K video, bufferbloat is a primary suspect. However, wireless airtime contention, ISP path congestion, or high baseline server latency produce identical symptoms. Always establish a quiet baseline on a wired Ethernet connection before attributing lag solely to router queueing.
Compare quiet and busy conditions
Keep the device, connection and test method consistent. Observe the affected application while the connection is otherwise quiet, then while a known transfer runs. Record both useful results and failures. Repeating a data-heavy test can use a metered allowance, so check your plan before running repeated comparisons.
Why new letter grades are withheld
Some testing tools, such as Waveform's Bufferbloat Test, assign letter grades based on added latency thresholds under load (such as requiring under +5 ms added delay for an A+ grade). In contrast, VelocityVerify estimates HTTP round-trip timing, not UDP or ICMP packet timing. It does not measure a separate upload-loaded latency series or end-to-end packet loss. New letter grades are withheld. See the measurement method and validity rules before interpreting a result.
A latency increase does not locate a queue
A repeatable increase during transfers is consistent with queueing along the route, but a browser test cannot pinpoint which device or physical link is holding the queue. Endpoint processing delays or wireless contention can also contribute. Where practical, compare a wired client directly at your router with your usual Wi-Fi setup to separate wireless contention from access-link buffering.
| Observation under active load | Likely queue location | Recommended next diagnostic action |
|---|---|---|
| Latency spikes on Wi-Fi, but stays flat on wired Ethernet | Local wireless hop (airtime contention or mesh backhaul) | Switch to 5 GHz/6 GHz band, reduce channel width, or use direct wired Ethernet before changing router queue rules. |
| Latency spikes during uploads only, on both wired and Wi-Fi | Upstream access buffer (cable modem, DSLAM, or router WAN queue) | Enable SQM on router egress, shaping upload speed to 85–90% of verified upload throughput. |
| Latency spikes during downloads only, on both wired and Wi-Fi | Downstream ISP buffer or router ingress interface | Enable SQM on router ingress, shaping download speed to 90% of sustainable download throughput. |
| Latency is high continuously, even with zero network traffic | Physical propagation distance, ISP peering, or remote server route | Do not change SQM; queue management cannot reduce baseline fiber-optic travel delay to distant endpoints. |
Record adapter and connection changes rather than treating Ethernet as a guarantee of a fault-free service.
Try supported SQM carefully
OpenWrt SQM documentation explains traffic shaping, active queue management, and scheduling, specifically recommending CAKE and FQ-CoDel (RFC 8290). CAKE is a shaping-capable queue discipline that combines flow isolation, active queue management and traffic prioritization to prevent single bulk streams from starving latency-sensitive traffic.
Annotated SQM provisioning calculation (The 85–90% rule)
SQM is configured with the goal of making your router the controlled bottleneck link in the path. SQM can reduce delay at a bottleneck it effectively controls, depending on conditions and setup. By capping throughput slightly below your link's sustainable capacity, packets can queue in your router's smart, fair queues rather than in an unmanaged modem or ISP buffer:
| Connection parameter | Measured sustained throughput | Target SQM rate setting (90% OpenWrt baseline) | Technical rationale |
|---|---|---|---|
| Download (Ingress) | 100 Mbps | 90,000 kbps (90 Mbps) | Packets arrive after traversing the access link; shaping and dropping or marking packets at the router encourages responsive senders to slow down, reducing upstream queue buildup over time. |
| Upload (Egress) | 20 Mbps | 18,000 kbps (18 Mbps) | Isolates outbound flows so bulk uploads do not starve interactive traffic; explicit priority for voice, gaming, or ACK packets depends on classification rules and qdisc features. |
On variable-speed connections (such as DOCSIS cable subject to peak-hour neighborhood congestion or fixed wireless with changing link rates), setting an SQM rate based on off-peak throughput will not prevent unmanaged ISP buffering if line capacity drops during busy periods. Instead, measure sustainable capacity during the hours when slowdowns occur, adjust shaping rates conservatively below that measured baseline, and validate latency under active load. Back up your configuration, identify the correct WAN interface, and test one setting at a time. Shaping trades a small amount of peak throughput for consistent responsiveness.
Check capabilities before buying
Check whether the exact hardware revision and firmware support the required controls, whether the device can process traffic at your intended rate, and whether offloading conflicts with queue management. OpenWrt documents CPU and hardware-flow-offloading limitations. A product name or QoS badge alone is not evidence of performance. This guide no longer ranks routers as lab-tested purchases.
Treat QoS labels as a starting point
Router interfaces use different names for prioritization, shaping and queue management. Consult the vendor documentation rather than assuming a gaming preset implements SQM. Avoid stacking several automatic optimizers while troubleshooting: if conditions improve or worsen, you need to know which change produced the difference.
Verify and keep a rollback path
Repeat the same observations after the change and compare both throughput and the original symptom. Include quiet and busy periods, and keep unsuccessful attempts in the record. Restore the earlier configuration if service becomes unstable or the trade-off is unsuitable. A better browser result is not a guarantee that every game server or call will behave the same way.