↑↓ to navigate Enter to open Esc to close

Bufferbloat Test: Latency Under Load, Graded A+ to F

By

Speed tests measure how fast your line is when nothing else is happening. This test measures what happens to your latency when your connection is actually busy under a saturated download or upload — the thing that ruins video calls and games.

Bufferbloat Test

Live
1Baseline
2Download
3Drain
4Upload
Latency —ms
Idle Download saturated Upload saturated Drain dashed line = idle baseline
Ready. The test takes about 40 seconds.
Advanced settings

Point these at your own server if you host one. "auto" picks the fastest reachable probe endpoint at test time.

The test transfers data at your full line speed — roughly 1–4 GB on a fast connection. Avoid running it on a metered or capped connection.

Self-contained test — no tracking, no analytics, no data retention. Load traffic uses Cloudflare's public speed-test endpoints. This tool provides estimates for planning purposes, not a guarantee of real-world performance.

About bufferbloat

Bufferbloat is latency caused by oversized buffers in your router or modem. When your connection gets saturated by a big upload, a cloud backup, or someone streaming, packets pile up in those buffers instead of being dropped promptly. Nothing is technically lost, but everything arrives late. An idle ping of 20 ms can jump to 300 ms or more, and real-time apps fall apart.

A comparison of a correctly-sized network buffer, where packets pass through quickly, against an oversized buffer, where packets queue up and arrive late Correctly sized buffer Packets pass through immediately — low, stable latency Oversized buffer (bufferbloat) Packets queue up waiting their turn — latency climbs the longer the queue grows Idle Loaded, small buffer Loaded, oversized buffer ~20 ms ~40 ms ~300+ ms
Bar heights are illustrative, not measured — the point is that latency under load depends entirely on how the buffer is sized, not on how fast the underlying connection is.

This test measures your latency three ways: at rest, while your download is fully saturated, and while your upload is fully saturated. The difference between idle and loaded latency is your bufferbloat — the delta matters more than the absolute number, since a naturally higher idle baseline (satellite, for instance) shouldn't be penalised the same way a small delta on a low-latency line would be. It is graded on the same A+ to F scale popularised by DSLReports, so you can compare the result with other tools directly.

GradeLatency increase under loadWhat it feels like
A+under 5 msPerfect. Nothing notices the load.
A5–30 msExcellent. Calls and games stay clean.
B30–60 msFine for most things, slightly loose under load.
C60–200 msNoticeable. Video calls stutter when someone uploads.
D200–400 msBad. Real-time apps break during transfers.
Fover 400 msSevere. The connection is unusable while busy.

Not every network complaint is bufferbloat. Before changing router settings, match what you are actually experiencing against the most likely cause:

What you noticeMost likely causeHow to confirm
Calls stutter only while someone uploads or backs upBufferbloat, usually on the upload sideThis test — compare the upload-loaded grade against idle
Everything is slow all the time, even when idleThroughput or line problem, not bufferbloatA plain speed test against your plan's advertised speed
Lag spikes at the same hours every eveningNeighbourhood congestion (common on cable)Run this test at a quiet hour and again at peak, compare
Only Wi-Fi devices suffer; wired ones are fineWi-Fi interference or signal, not the router queueRun this test twice: once on Ethernet, once on Wi-Fi
Short freezes where nothing moves, then a burstPacket loss rather than queueing delayPacket Loss Calculator and this test's timed-out probe count

Common mistakes

Testing while the household is already using the line, then blaming the grade. A backup, a game download, or someone streaming during the baseline phase inflates your idle reading and shrinks the measured delta — the grade comes out better or worse than reality at random. Quiet network first, normal-traffic run second.
Comparing a Wi-Fi grade against an Ethernet grade as if they measure the same thing. Wi-Fi adds its own queueing and retransmissions on top of the router's. A B on Ethernet and a D on Wi-Fi is not a contradiction — it tells you where the problem actually lives.
Setting the SQM bandwidth cap at or above the line's real speed. SQM only works when the bottleneck queue sits inside your router. Cap it above your actual throughput and the queue stays in the modem or the ISP's equipment, unmanaged — the setting is on, the grade does not move.
Judging from a single run. One badly-timed neighbourhood spike or a background app update can swing one run by a full letter grade. Run it two or three times and trust the pattern, not the outlier.

How to fix bufferbloat

  1. Enable Smart Queue Management (SQM). Modern queue algorithms — fq_codel and CAKE — keep buffers short by design. Look for "SQM", "QoS", or a gaming/conferencing optimisation setting in your router. On eero, enable "Optimize for Conferencing and Gaming" in Labs. On OpenWrt, install the luci-app-sqm package and set your bandwidth to about 90–95% of your measured line speed.
  2. Cap the shaper slightly below your real speed. SQM only controls the queue if the bottleneck is inside your router. Setting the limit a little under your actual throughput moves the queue where you can manage it.
  3. Update or replace old equipment. Routers and modems from before roughly 2018 rarely have modern queue management. If your grade is D or F and your router has no SQM option, the router is usually the fix.
  4. Re-test after every change. Toggle a setting, run the test again, compare grades. SQM done right typically turns an F into an A.

Where the queue-management setting lives depends on the router. The names differ, but they all do the same job:

Router / firmwareSetting nameWhere to find it
OpenWrtSQM with cake or fq_codelInstall luci-app-sqm, then Network, then SQM QoS
eeroOptimize for Conferencing and Gamingeero app, then Settings, then eero Labs
ASUSAdaptive QoS (fq_codel on recent firmware)Adaptive QoS, then QoS, then Gaming or Adaptive mode
Ubiquiti UniFiSmart QueuesInternet settings for the WAN connection
FirewallaSmart QueueNetwork manager, per-WAN Smart Queue toggle
MikroTikCAKE / fq_codel queue typesQueues, requires manual configuration
Typical ISP-supplied boxUsually nonePut it in bridge mode and run SQM on your own router behind it

Menu locations move between firmware versions — if a path above does not match your interface, search the router's manual for the setting name instead.

How this test works

The latency probe is a tiny HTTP request sent every 100 ms to a different origin than the one serving the load traffic, so it travels on its own connection instead of queueing behind the bulk transfer inside the browser. Load traffic is served by Cloudflare's public speed-test endpoints. Nothing is stored — the whole test runs in this page.

Known limitation. A probe over TCP/HTTP is not quite as clean as a dedicated UDP flow. If your browser or OS is the bottleneck rather than your router — which happens on multi-gigabit lines, above roughly 5 Gbps — the download result can understate real bufferbloat. In practice that is rarely a concern, because bufferbloat almost never shows up on a connection that fast.

Run the test with everything else on your network idle for a clean reading, then run it again with your normal traffic going to see what your household actually experiences.

Frequently asked questions

What is bufferbloat, in plain terms?

Your router queues packets before sending them. If that queue is too large, packets sit and wait instead of being sent promptly or dropped — so everything arrives late during busy moments, even though nothing was technically lost.

Why does my speed test look fine but calls still lag?

A standard speed test measures throughput when the line is otherwise idle. Bufferbloat only shows up when the connection is actually busy — exactly the condition this test creates on purpose.

Is this test safe to run on a metered connection?

No — the load phases transfer roughly 1–4 GB on a fast connection to genuinely saturate it. Run it only on an unlimited or generously capped plan.

Why is upload usually worse than download?

Most residential connections have far less upload capacity than download, so it saturates faster and the router's upstream buffer fills more easily — which is why cable and DSL connections in particular often grade worse on upload.

Does a mesh Wi-Fi system affect the result?

Yes — Wi-Fi itself adds its own queueing and can mask or worsen router-level bufferbloat. For the cleanest reading of your router/modem specifically, run the test from a device connected by Ethernet if possible.

Will running this test on a VPN change the result?

Yes, usually for the worse — VPN encryption and routing add their own latency and can introduce queueing at the VPN server. Disable it for a baseline reading of your own network.

Does enabling SQM reduce my speed?

Yes, slightly — capping the shaper below your line rate typically costs about 5–10% of raw throughput, and cake/fq_codel add a little CPU work on the router. For almost everyone that trade is worth it: a line at 90% speed that stays responsive under load feels far faster in daily use than a line at 100% that collapses whenever it is busy.

What happened to the DSLReports bufferbloat test?

DSLReports Speed Test, the tool that popularised the A+ to F bufferbloat grade for over a decade, shut down in 2025 after its operator wound down the site's aging infrastructure. This test uses the same grading scale so results remain directly comparable to any DSLReports screenshot you may still have saved.

Sources

  • Bufferbloat problem description. Bufferbloat.net, maintained by the Bufferbloat project (Dave Täht, Jim Gettys, and contributors).
  • A+ to F grading scale. Popularised by DSLReports' bufferbloat test (2015–2025), now a de facto standard for comparing results across tools.
  • fq_codel and CAKE queue management. IETF RFC 8290 (FlowQueue-CoDel) and the CAKE traffic shaper documentation.

Related calculators

See the full list of Networking tools, or try: