Fast fibre can still feel terrible when a download or cloud upload fills oversized queues. Bufferbloat in South Africa is the repeatable increase in latency under load, not simply a high ping to a distant server. Smart Queue Management moves the bottleneck to a router that can control it, but only after a wired test proves congestion is the cause.

Quick Answer

Measure idle ping, then repeat while saturating download and upload over Ethernet. If latency rises sharply only under load, configure SQM slightly below the sustained line rates and start with CAKE, or fq_codel when router CPU is limited. The software may cost R0 on compatible OpenWrt hardware; buy a new router only if the current CPU cannot shape the connection speed.

bufferbloat South Africa decision map

📏 Build a Repeatable Baseline

Connect one computer by Ethernet, pause other traffic and record idle latency to the router and a stable internet destination. Then run a sustained download while ping continues. Repeat with a sustained upload, because upstream queues often reveal the problem more clearly. Save minimum, average, maximum and visible spikes rather than relying on one speed-test grade.

Do the test more than once and at different times. A latency increase that follows your own saturation is consistent with bufferbloat. Packet loss that begins before load, unstable first-hop Wi-Fi or a route that is always slow points elsewhere. Use wired networking equipment for the baseline so radio interference is not mixed into the result.

🌊 Why SQM Works

Without shaping, the slowest link may sit inside an ISP device or network terminal where your router cannot manage the queue. SQM deliberately makes the router's configured rate slightly lower, causing packets to queue locally. CAKE or fq_codel can then share capacity between flows and signal congestion before delay grows excessively.

OpenWrt recommends CAKE as the comprehensive default and describes fq_codel as faster but less comprehensive. Neither creates bandwidth. Both trade a small amount of peak throughput for controlled latency and fairer flow scheduling.

⚙️ Configure the Correct Interface and Rates

Install only the supported SQM packages for the router's OpenWrt release. Select the WAN interface that carries the internet traffic, not a guessed Ethernet label. On PPPoE systems it may be a logical interface. If the wrong interface is shaped, the speed changes little and loaded latency remains high.

Measure actual sustained throughput rather than copying the package label. Start with download and upload rates modestly below those measurements, apply settings, and repeat the exact test. Reduce the direction that still shows spikes in small steps. If both latency and throughput collapse, confirm units, interface and CPU use.

OpenWrt's guide uses Ethernet link-layer adaptation for fibre and notes that overestimating overhead is safer when uncertain. Provider encapsulation differs, so preserve the original settings and change one value at a time.

Readers choosing compatible hardware can compare fibre routers, but should verify OpenWrt support and routing performance for the exact revision before ordering.

🧠 Check Whether the Router CPU Is the Limit

SQM processes traffic on the CPU. A router that handles basic gigabit forwarding may not run CAKE at the same speed. During the loaded test, watch CPU utilisation and the shaped throughput. If a core is saturated and speed falls far below target, try fq_codel or a simpler configuration before replacing hardware.

Hardware flow offload can bypass shaping on some configurations. Follow the OpenWrt guidance for the installed release and verify that the queue statistics increase during traffic. A configuration page showing "enabled" is not proof that packets traverse the discipline.

🎮 Validate With the Real Gaming Household

After the synthetic test passes, play while another device uploads or downloads. Look for stable frame-to-frame network behaviour, voice-chat quality and reduced latency spikes. SQM should help multiple flows coexist; it cannot shorten the physical route to Europe or repair loss inside an ISP backbone.

Keep the before-and-after numbers, firmware version, interface, rates and queue discipline. Fibre profiles can change after an upgrade, and the old shaping values may then be too high or unnecessarily low. Retest whenever the service rate changes.

Test upload and download independently after tuning. A household can have clean download latency but poor upload behaviour during cloud backup, so one combined score may conceal the direction that still needs a lower shaping rate.

Keep test traffic on one wired client for the first comparison, then add the normal household devices. This separates queue behaviour from Wi-Fi airtime and shows whether per-flow fairness holds when a console, television and backup job compete. If one device still dominates, verify that its packets actually pass through the SQM router rather than another gateway.

Repeat the check after every major firmware update.

The networking category includes broader equipment, but spend only after the CPU ceiling, port speed and software compatibility are known.

🧯 Know What SQM Cannot Fix

Weak Wi-Fi needs channel, placement or access-point work. Constant loss to the router suggests cable, port or local hardware trouble. High international latency is constrained by route distance. CGNAT affects inbound reachability, not queue length. Treating all four as bufferbloat leads to endless tuning.

Revert SQM temporarily and repeat the baseline if results become confusing. A clean A/B test is more useful than an elaborate priority rule set. OpenWrt's modern queue disciplines are designed to work without manually classifying every game port.

Frequently Asked Questions

How do I test fibre bufferbloat?

Compare wired idle ping with ping during sustained upload and download. A large, repeatable increase under your own load is the signature.

Is CAKE better than fq_codel?

OpenWrt prefers CAKE for comprehensive mitigation. Fq_codel can preserve more throughput when router CPU is the limiting factor.

What SQM rate should a 100Mbps line use?

Measure sustained wired throughput first and configure slightly below that result. Package labels and real rates are not always identical.

Will SQM lower download speed?

Usually by a controlled margin. That margin lets the router manage the queue instead of allowing delay to build upstream.

Can SQM fix overseas server ping?

It can remove local congestion delay, but it cannot change geographic distance or upstream route selection.

Why does SQM slow down a gigabit line too much?

The router CPU may be saturated, the wrong units may be entered, or acceleration and interface selection may be incorrect.

Ready to trade a little peak speed for stable play? Capture the wired baseline, tune one direction at a time, and choose new routing hardware only after CPU measurements prove it is necessary.