Your internet isn't slow — it's bufferbloat. Diagnose it in 25 seconds.

You pay for 500 Mbps. At 2 a.m. everything is instant. At 7 p.m., when someone starts a game update, your video call turns into a slideshow — and the speed test you run right after still cheerfully reports 500 Mbps.

Nothing is broken. Your speed test just doesn't measure the thing that's actually hurting you.

The number that speed tests don't show you

Raw bandwidth answers "how much data can move per second". It says nothing about how long each tiny packet waits when the line is busy — and that latency-under-load number is what real-time apps live and die by.

That problem has a name: bufferbloat.

Modern routers and modems buffer packets instead of dropping them. That's polite under light load, but when a download saturates your link, outgoing packets — your video call frames, your game inputs — get stuck at the back of a huge queue. Your ping goes from 8 ms to 300+ ms while the download runs. The call stutters. The game desyncs. When the download finishes, everything is magically fine again, which is why it feels haunted.

Jitter is the sibling problem: the variation in your ping between packets. Real-time codecs can hide high latency better than they can hide inconsistent latency.

The 25-second diagnosis

You need a test that measures latency while the line is saturated, not just at idle. Waveform's bufferbloat test is the well-known option.

SpeedPulse is the one I built — free, no signup, no app, no cookies, everything runs client-side. One ~25-second run measures:

• Download / upload / idle ping — the usual suspects
• Jitter (RFC 3550-style mean difference between consecutive pings)
• Bufferbloat — a ping train keeps running during the download and upload phases; the tool reports loaded latency − idle latency per direction and grades the worse one from A+ to F
• Plain-English verdicts for 4K streaming, gaming, video calls and browsing, so you don't have to interpret raw numbers
• An 8-CDN cross-check — after each test it pings 8 independent CDN edges (Cloudflare, jsDelivr, Fastly, GitHub, Google, AWS...). If the speed-test server is dramatically faster than the pack, your ISP may be prioritizing speed-test traffic — the classic reason "the test says I'm fine but everything feels slow"
• CSV export and a local history chart, all client-side

Reading the grade

If you score C or worse, you've found your problem.

Full disclosure: I built SpeedPulse. The grade scale is the same A+–F scheme Waveform and the bufferbloat.org community use, so results are comparable across tools — I'd rather share a standard than invent a flattering one.

The fixes that actually work

In order of effectiveness:

1. Enable SQM with fq_codel/CAKE on your router (trivial on OpenWrt; many stock firmwares ship a basic "QoS"). Smart queue management keeps buffers short so latency stays flat even at full load. This is the fix — A+ grades come from this.
2. Cap your speeds ~5–10% below your ISP rate in the SQM settings. Counter-intuitive but correct: queues then form inside your router, where SQM controls them, instead of inside the ISP's modem, where they don't.
3. Fix the upload direction first. Skinny upstreams bloat first — video calls die on upload saturation (cloud backup + call = slideshow). If you can only cap one side, cap upload.
4. Go wired for the machine that matters. Wi-Fi retransmissions and airtime contention add jitter no router setting can undo.
5. Check for speed-test favoritism. If only the test's server is fast while every other CDN is slow, the 8-CDN panel makes the mismatch visible — worth raising with your ISP.

For the nerds: how you measure this in a browser

No ICMP, no raw sockets. A browser test times HTTPS round trips with fetch(url, { mode: 'no-cors' }) and performance.now(). no-cors returns an opaque response — no body, no status — but the full network round trip (DNS, TCP, TLS, HTTP) completes, which is all timing needs. Jitter is the mean absolute difference between consecutive samples; bufferbloat is the median loaded latency minus the idle median, measured per direction, graded on the worse of the two.

Full methodology write-up, including the limitations we're honest about: How SpeedPulse measures jitter, bufferbloat and 8 CDN latencies.

Run the test on your line and drop your grade in the comments — happy to help interpret results: speedpulse.tech. Want the long-form bufferbloat explainer (router queue theory, CAKE vs fq_codel)? What is bufferbloat?