Your ping sits at 15 ms all evening. Then someone in the house starts a Twitch stream or a phone kicks off a photo backup, and your next fight rubber-bands at 200 ms. Nothing about your game changed. The connection did. This is bufferbloat, and it comes from how your router handles a full pipe, not from how fast that pipe is.
Why one upload wrecks everyone’s ping
Home plans are lopsided. You might have 500 Mbps down and 20 Mbps up. Games barely touch either figure. A match sends tiny state updates, a few hundred kilobits a second. So the big download number on your bill has almost nothing to do with whether you lag.
The upload pipe is the weak point. Cloud photo sync, a game upload, or OBS pushing 8 Mbps to Twitch fills 20 Mbps of upload in about a second. Once that pipe is full, your router has more packets than it can send, so it parks the overflow in a buffer.
That buffer is the trouble. Consumer modems and routers ship with buffers sized for throughput, not for latency. They would rather hold a big backlog than drop a single packet. Your game’s next input joins the back of that line and waits behind a second of bulk data. The wait is your spike.
An upload flood does something worse: it slows traffic coming back the other way too. Downloads depend on small acknowledgment packets your machine sends back upstream. When the upload queue is jammed, those acknowledgments are stuck in it too, so the whole connection drags in both directions from one saturated uplink.
Confirm it is bufferbloat, not something else
High ping has more than one cause. A distant game server, a weak Wi-Fi signal, or a bad route all look like lag from your chair. Bufferbloat has a signature: your ping is fine at rest and only climbs when the line is busy.
The bufferbloat test measures exactly that. It clocks your idle ping, saturates the download and upload, then clocks the ping again under load. A small gap earns an A. A ping that doubles or worse earns a D or F. If your idle number is healthy and the loaded number balloons, bufferbloat is your answer. If your idle ping is already high, the cause is elsewhere, usually server distance or Wi-Fi. Run the jitter test to check for instability on an idle line.
The fix: shrink the queue, do not widen the pipe
A faster plan doesn’t help. More bandwidth just gives the buffer more data to hoard. The fix is Smart Queue Management, or SQM, a router feature that keeps the queue short and routes latency-sensitive packets around the bulk transfer.
SQM runs one of two modern algorithms. fq_codel watches how long packets sit in the queue and drops or marks them before the backlog grows, while giving each traffic flow its own fair lane. CAKE is the newer option. It does the same queue control, adds per-device fairness so one machine’s upload can’t starve the rest of the house, and handles the shaping in a single step. If your router offers both, pick CAKE. Either one beats a router with no queue control at all.
Where you actually get SQM
Not every router can do this, and this is where most guides go quiet.
- OpenWrt. The free router firmware ships an SQM package (
luci-app-sqm) built for this exact job. Install it on a long list of consumer models, and it’s the most dependable route of the three. - Some stock firmware. A few off-the-shelf routers expose real SQM or proper shaping. ASUS models running Merlin firmware, certain FRITZ!Box units, and Ubiquiti EdgeRouters can do it. The generic “QoS” slider on a cheap router usually can’t: it reorders priorities without controlling the queue, so it does little under a real flood.
- Your ISP gateway, usually not. The combined modem-router your provider hands you rarely runs SQM, and you often can’t change it. Bridge mode is the workaround: set the gateway to pass traffic straight through, then put your own SQM-capable router behind it.
Set the cap right
SQM only works when your router is the bottleneck, not the ISP. So you cap your rates just below your real line speed. Measure your actual download and upload first, then set the SQM limits to about 85 to 90 percent of those numbers. If your line measures 100 down and 20 up, cap around 88 and 17.
You give up a slice of peak throughput on paper. In return your loaded ping falls from hundreds of milliseconds to single digits, which is the trade you want for gaming and calls. Set the cap too high and the queue slides back to the ISP where SQM can’t reach it, and the spikes come back.
When the router cannot fix it
Some spikes are not yours to solve. Cable connections share neighborhood capacity, so a congested segment at peak hours adds latency no home router can touch. If your loaded ping stays bad after SQM is tuned and confirmed, and it worsens at the same time every night, the bottleneck is upstream. That is an ISP conversation, and sometimes a reason to price out a fiber provider.
Two quick wins are worth doing no matter what. Wire your gaming device with Ethernet, since Wi-Fi adds its own variable delay on top of any bufferbloat. And pause the obvious offenders during a session: cloud backups, system updates, and other uploads are the usual trigger.
Prove the fix worked
Don’t trust the settings blind. Run the bufferbloat test before you touch anything and note the grade. Enable SQM, set your caps, then run it again. A jump from an F to an A, with your loaded ping now near your idle number, is the proof. If the grade didn’t move, your cap’s probably too high, or your router isn’t truly shaping the queue.