What Is Bufferbloat? How to Test and Fix It in 2026
HIGH PING UNDER LOAD • LOADED LATENCY • SQM • CAKE • FQ_CODEL
Bufferbloat is the extra latency that appears when a network bottleneck builds an oversized packet queue
The classic symptom is simple: your connection feels fine while idle, then your ping shoots up when somebody starts a large download, cloud backup, upload or speed test.
It can happen on slow broadband or very fast Full Fibre. The important question is not just how many Mbps you have; it is how well the bottleneck queue is managed when the connection is busy.
What bufferbloat looks like
Bufferbloat.net defines bufferbloat as undesirable latency caused by excessive buffering in routers or other network equipment. It is most obvious when the bottleneck is saturated.
If you are trying to separate different connection problems, our guide to packet loss vs high ping explains why similar symptoms can have different causes.
It is not always the router
The troublesome queue forms at the bottleneck: the slowest point that is currently limiting the flow. That may be the WAN connection, an ISP device, a modem or even Wi-Fi.
Waveform’s current test guidance makes this distinction explicitly. If your broadband is gigabit but a Wi-Fi link can only move around 100Mbps in that location, the wireless link can become the active bottleneck and show its own loaded-latency problem.
For that reason, test over Ethernet first when possible. Then repeat over Wi-Fi if you want to understand whether the wireless link adds another bottleneck. See our Wi-Fi vs Ethernet guide for the practical differences.
How to test for bufferbloat properly
The Waveform Bufferbloat Test is useful because it measures idle latency, then measures latency again during active download and upload. Its grade is based on the increase in latency under load rather than your raw broadband speed.
| Result | What it suggests | What to check next |
|---|---|---|
| Idle ping is good, loaded ping jumps | Classic sign of queueing delay at the active bottleneck. | SQM, bandwidth shaping and where the bottleneck actually sits. |
| Idle ping is already poor | Bufferbloat may not be the main issue. | Routing, mobile/satellite link conditions, Wi-Fi, line fault or server distance. |
| Ethernet is good, Wi-Fi is poor | The wireless hop may be the limiting queue or adding interference. | Access-point placement, Wi-Fi airtime, client link rate and wireless queue management. |
| Upload causes the biggest spike | Upstream saturation is the main pain point. | Prioritise getting upload shaping under control first. |
Before changing anything, it also helps to test your internet speed properly so your SQM limit is based on a realistic connection rate.
The main fix: Smart Queue Management
Smart Queue Management (SQM) combines traffic shaping, fair queueing and active queue management so the router can keep the bottleneck queue short instead of allowing packets to pile up.
FQ-CoDel is formally described in RFC 8290. CAKE builds on years of SQM deployment experience and integrates more of the shaping and fairness logic into one system.
Why SQM usually sacrifices a little headline speed
The goal is to make your router’s controlled SQM queue become the bottleneck before a less-manageable queue forms in the modem or ISP equipment. That is why you deliberately leave some throughput on the table.
Do not treat 90% as a sacred number. Connections with stable provisioned rates may work closer to full speed; variable-rate links may need more headroom or dynamic shaping. Start with a known baseline, enable SQM, retest loaded latency and tune from there.
A successful result is not necessarily the biggest speed-test number. It is a connection that stays responsive while it is being heavily used.
Can your router actually run SQM at your full speed?
This matters more as broadband gets faster. Queue management consumes CPU resources, and an older or low-powered router may control bufferbloat well at 100Mbps but fail to shape a gigabit line without reducing throughput heavily.
Current OpenWrt guidance describes CAKE as the preferred queue discipline for bufferbloat, while noting that fq_codel can achieve higher throughput on some CPU-limited hardware because it is less comprehensive.
Can you fix bufferbloat without buying a new router?
Often, yes. If your current router already supports reliable SQM or bandwidth shaping, configuration may be enough. The useful sequence is:
- Measure realistic upload and download speed with the connection otherwise idle.
- Enable SQM on the actual internet-facing interface.
- Start around 90% of the measured rates.
- Retest loaded latency over Ethernet.
- Raise or lower the limits gradually until you get the best balance of throughput and responsiveness.
If the router has no useful queue-management feature, or cannot process your line speed with SQM enabled, then replacement hardware becomes more relevant.
The bottom line
Bufferbloat is not “slow broadband”; it is excessive latency when a bottleneck queue fills. Test idle versus loaded latency first. If the problem is real, SQM with CAKE or fq_codel can keep queues short by deliberately controlling the bottleneck, often starting at around 90% of measured line speed. The best result is a connection that stays responsive while somebody else is using it heavily.
Bufferbloat FAQs
Can bufferbloat happen on a fast Full Fibre connection?
Yes. Bufferbloat is about excessive latency when a bottleneck queue fills, not simply about having a slow connection. A fast line can still suffer under load if the bottleneck is poorly managed.
Should I set SQM to exactly 90% of my broadband speed?
OpenWrt recommends 90% of a measured baseline as a sensible starting point. It is not a universal magic number: retest after enabling SQM and adjust if the queue is not controlled or if too much throughput is being sacrificed.
Is CAKE better than fq_codel for bufferbloat?
Both are designed to control latency under load. Current OpenWrt guidance describes CAKE as the preferred, more comprehensive queue discipline, while fq_codel can be lighter on router hardware and may achieve higher throughput when CPU performance is the limiting factor.
Sources checked
- Bufferbloat.net: More about Bufferbloat — definition of bufferbloat and why modern queue-management algorithms reduce latency.
- OpenWrt: Smart Queue Management — current SQM setup guidance, 90% starting rates and CAKE versus fq_codel notes.
- OpenWrt: SQM configuration — current queue-discipline and shaping configuration details.
- Waveform Bufferbloat Test — current loaded-latency test methodology and grading approach.
- RFC 8290: FQ-CoDel — technical specification for the FQ-CoDel packet scheduler and active queue-management algorithm.