Why your ping spikes when the connection is busy
The pattern is always the same: latency is fine until something starts downloading, then it jumps by hundreds of milliseconds for a couple of seconds and settles again. Calls stutter, games rubber-band, and every speed test insists the line is healthy. That is bufferbloat — a queue somewhere on the path holding your packets while it drains a bulk transfer.
Three steps that name the culprit
- Measure idle latency first. With nothing else using the network, ping a stable address for a few seconds and note the median. This is your baseline — without it the second number means nothing.
- Measure again under load. Saturate the line with a download, then with an upload, and keep pinging throughout. The rise over your baseline is the bufferbloat: under 30 ms is healthy, over 200 ms is what breaks calls.
- Read the direction. Latency that explodes while uploading points at the buffer in your own router or modem. Latency that explodes while downloading points at your provider's equipment. The two have completely different fixes.
What the grade means
The grade is how far latency rises while the line is loaded, measured against your own idle baseline — not the baseline itself, and not your speed. A 40 ms line that stays near 40 ms under load grades better than a 10 ms line that jumps to 150. NetDiag+ takes the 95th percentile of that rise in the worse of the two directions, so one stray packet does not decide the result.
- A+ — under 5 ms of rise. The queue is effectively empty.
- A — under 30 ms. Healthy; most people never notice anything.
- B — under 60 ms. Still fine for calls; competitive games may feel it.
- C — under 200 ms. Video calls start to stutter whenever someone else downloads.
- D — under 400 ms. Calls drop out and games become unplayable under load.
- F — 400 ms or more. The classic “the internet dies whenever anyone uploads photos”.
Wi-Fi, router or provider: reading the split
A single ping under load proves that a queue exists, not where it is. NetDiag+ pings two targets throughout the test — your router and a host on the internet — and the way the two behave places the queue:
- Router latency climbs too (by more than about 30 ms): the queue sits between your phone and the router — Wi-Fi congestion, a weak signal or the phone’s own radio. Test again next to the router, or from another device, before you change any setting.
- Router flat, upload at least twice as bad as download: the uplink buffer in your own router or modem. This is the case that SQM fixes.
- Router flat, download at least twice as bad as upload: the queue is in your provider’s network. You cannot fix it at home, but you can document it.
- Router flat, both directions bloated: the modem, or the line between it and the provider.
- On mobile data the queue is in the carrier’s network by design, and the result says so instead of blaming anything in your home.
When the test cannot load the line
Every bufferbloat test assumes it actually filled the line. If the test server never delivers real data, latency stays flat, the rise is close to zero and a naive test hands out an A+ for a line it never tested. This is more common than it sounds: since June 2025, connections from Russia to Cloudflare are throttled so that only about the first 16 KB of each response arrives, and captive portals and corporate proxies can stall a download in a similar way.
So always read the speeds next to the grade. If neither direction reached about 1 Mbps and latency did not move, the run says nothing about your line. NetDiag+ reports such a run as ungraded — the test server could not be loaded from this network — instead of a clean grade, and Health Score leaves its load and speed points neutral rather than counting them as perfect.
What actually fixes it
- Upload bloat: turn on SQM. On routers that offer it, enable Smart Queue Management (CAKE or fq_codel; vendors also call it QoS or Adaptive QoS) and set the upload limit to roughly 85–95% of the speed you actually measure. The router then keeps the queue in its own hands instead of letting the modem fill a huge buffer.
- Download bloat: shape below the line or escalate. Setting the download limit a little under your real speed often softens it. If it does not, the queue is upstream, and a set of daytime and evening runs is the evidence your provider needs.
- Wi-Fi bloat: fix the radio, not the shaper. Move closer, switch to 5 GHz, change the channel or add an access point. SQM on the router cannot shorten a queue that forms in the air.
- Re-test at the bad hour. A fix that grades A at noon has proved nothing until it holds up in the evening, when the line is busiest.
Free · 31 tools · one-time purchase · no subscription
Questions people actually ask
My speed test says 500 Mbps — why do I still lag?
Because a speed test measures how much data fits through an empty pipe, not how long one packet waits when the pipe is full. Bandwidth and latency-under-load are different measurements, and only the second predicts whether a video call survives.
Is this my router's fault or my provider's?
The direction tells you. Upload bloat is almost always the buffer in your own router or modem, because that is the device choosing what to send next. Download bloat sits in your provider's equipment, because they choose what to send you.
Why did my provider say everything is fine?
Their tests check the line: signal levels, error counters, throughput. All of those can be perfect while a queue adds 300 ms of delay. You are not being fobbed off — they are measuring something else.
Can I fix it myself?
If the bloat is on upload, usually yes: enable SQM, QoS or fq_codel on your router and cap the upload at roughly 90% of your line speed. That deliberately trades a little bandwidth for a short queue. Download bloat you can only soften by rate-limiting below your line speed.
My phone tests clean but my PC still lags. What now?
Then the network is not the problem and the fault is in that machine. Network adapter drivers are the usual suspect: generic in-box drivers ship with oversized transmit buffers and no queue management, which produces exactly these spikes. A VPN client or security suite can do the same.
Is one test enough?
No. Bufferbloat is a time-of-day problem: a clean run at 3 am proves nothing about 9 pm. Run the test at the bad hour on a few different days and once at a quiet hour as a control. NetDiag+ keeps the worst spike of every run on a 7-, 30- or 90-day trend chart, coloured by grade, can remind you once to check again at the other end of the day, and builds a text or PDF report for your provider from your Health Score history.
Why do I get an F on my phone but a B on the laptop?
Because the phone test measures your line plus its own Wi-Fi radio. A phone adds a queue of its own, drops into power-save between packets and shares the air with every neighbour — so latency under load over Wi-Fi is usually worse than the line itself, sometimes by a whole grade. Neither result is wrong; they measure different things. NetDiag+ pings your router separately during the test: if the gateway already wobbles, the report says the local link is the problem and stops you blaming the provider for your Wi-Fi.
Bufferbloat is not the only thing that quietly breaks a healthy-looking connection. Next: how to find your real Path MTU →