Picture a plan sold as gigabit and a game holding a steady 15 ms ping. Then a phone in the next room starts pushing its camera roll to iCloud. Ping leaps past 300 ms, voice chat goes robotic and your character rubber-bands into a wall. The line isn't broken, and nobody is throttling you. The upload filled a buffer inside your cable modem. A pricier plan won't drain it. A free router setting that costs you about a tenth of your speed will.
Why does your game lag the moment someone in the house starts a video call or an upload?
The tell is the pattern. Ping and jitter look fine on a quiet network, then fall apart the moment anyone sends data out: a phone backing up to iCloud or Google Photos, a Zoom or Teams call, a Twitch stream, one fat email attachment.
That is bufferbloat. When a link is full, the box in front of it should drop a few packets early so senders slow down. Instead, oversized buffers hold everything in one long line. Your game's tiny packets wait behind megabytes of vacation photos, and latency climbs from tens of milliseconds to hundreds. As of 2026, a saturated 20 Mbps upload on an unmanaged router can add an estimated few hundred milliseconds of delay.
Jitter is the nastier half. The queue fills and drains unevenly, so delay swings from packet to packet. Calls break up and games stutter with zero packets lost, which is why a plain speed test swears everything is fine.
So if the line is fast, why does it choke?
Why doesn't a faster cable plan fix bufferbloat?
Because what you pay for is mostly download, and the jam is on the upload. As of 2026, on standard cable plans the upload is often an estimated tenth of the download or less. As of 2025, gigabit cable tiers from large US providers commonly paired about 1 Gbps down with roughly 35 Mbps up, varying by provider and market. As of 2026, mid-split and high-split upgrades are reaching some markets, but many homes still have thin uploads.
The buffer in front of that upload lives in your cable modem. As of 2025, moving from a 500 Mbps to a 1 Gbps cable tier often changed upload only slightly.
Cable engineers know this. As of 2026, DOCSIS 3.1 modems are specified to support PIE active queue management, described in RFC 8034 (2017), targeting roughly 10 ms of queuing delay. Uploads still bloat anyway:
- As of 2026, many older DOCSIS 3.0 modems do not have PIE.
- PIE manages average delay, not fairness between flows, so one bulk upload still drags everyone else down.
- Download queuing happens at your provider's CMTS, at the far end of the coax, where you can't touch it.
As of recent FCC Measuring Broadband America reports, latency under load is tracked, and cable has generally shown more added delay than fiber.
Our take: being told to upgrade your plan to fix lag is the trap cable companies profit from. As of 2025, the faster tier added download while upload often stayed around 35 Mbps. Shaping on your own router fixes what the upgrade won't, and it costs nothing.
Where the queue builds on a cable line, and where SQM moves it
- Phone or PCSends packets as fast as the network takes them
- Wi-Fi router without SQMPasses everything through at full speed, no queue control
- Cable modemThe upload queue piles up here and adds hundreds of ms
- Cable node and CMTSThe download queue forms here, out of your reach
- Router with SQM onCaps itself just below line speed, so the queue moves into the router and stays short
Is it bufferbloat or packet loss, and what does 42% or 100% loss mean?
Test before you touch a setting, on a wired computer:
- the Waveform bufferbloat test, which grades A+ to F by latency added under load
- Cloudflare's speed test, which reports loaded latency
- Ookla Speedtest, which shows download and upload latency separately
- the networkQuality command built into macOS since Monterey (released in 2021)
High added latency with zero loss is bufferbloat, and SQM is the fix. The grade thresholds are in the strip below.
Loss is a different animal. Sustained loss like 42%, or 100%, in your own ping to the router or to 8.8.8.8 means a line, Wi-Fi or hardware fault. Open the modem's status page (often 192.168.100.1 on many cable modems) and check the signal. As of 2026, commonly cited healthy cable signal ranges are downstream power about -7 to +7 dBmV, upstream about 35 to 50 dBmV, and downstream SNR around 33 dB or higher. Then hunt for loose coax, extra splitters or a dying modem, and call your provider for a line check.
One more sorting rule. A Netflix stream that keeps buffering is usually a throughput or Wi-Fi problem. Lag spikes in games and calls while someone else is busy are what SQM was built for.
What is SQM, and should you use fq_codel or CAKE?
Smart Queue Management makes your router throttle itself to slightly below your line speed on purpose. The queue moves out of the modem, where nothing manages it, into the router, where a smart queue discipline keeps every flow's delay short and splits the link fairly.
You have two real choices:
- fq_codel gives each flow its own small queue and drops or marks packets when a queue's delay stays high. It's light on CPU and the default in many Linux-based routers.
- CAKE is the newer all-in-one. It shapes, shares fairly per device (one gamer isn't drowned out by a sibling's ten download threads) and handles link overhead itself, DOCSIS preset included. It wants more CPU.
As of 2026, SQM typically costs an estimated 5 to 15% of peak throughput. Many older consumer routers can't shape much past a few hundred Mbps, so CAKE on a gigabit plan needs real hardware.
And the blunt part: "gaming QoS" that only ranks apps, such as stock ASUS Adaptive QoS or Netgear Dynamic QoS, does not fix bufferbloat on its own. Here is how the three compare.
CAKE vs fq_codel vs gaming QoS
CAKE
All-in-one shaper with per-host fairness and a built-in DOCSIS overhead preset.
Pros
- Keyword 'docsis' sets overhead 18 and mpu 64 in one go
- Per-host fairness stops one device's many downloads from crowding out the rest
- Usually the best bufferbloat grades
Cons
- Heavier on CPU
- Older consumer routers can't shape near-gigabit speeds with it
Best for: OpenWrt, Asuswrt-Merlin, Firewalla or x86 routers with CPU headroom
fq_codel
Light flow-queuing discipline, the long-standing default in Linux-based routers.
Pros
- Low CPU cost
- Widely available (UniFi Smart Queues, simple.qos on OpenWrt, pfSense limiters)
- Big improvement over no SQM
Cons
- Overhead has to be set by hand
- Fairness is per flow, not per device
Best for: Weaker routers, or fast plans where CAKE runs out of CPU
App-priority 'gaming QoS'
Stock features like ASUS Adaptive QoS or Netgear Dynamic QoS that rank app traffic.
Pros
- One-click setup
- Can help on a router that is itself the bottleneck
Cons
- Doesn't move the queue out of the cable modem
- Doesn't control the upload buffer that causes the lag
Best for: Not a bufferbloat fix on its own
Bottom line: Use CAKE if your router can handle it, use fq_codel if it can't, and don't count app prioritization as SQM.
What download and upload limits should you set for your cable speed?
Start from what your line delivers, not the plan name. Cable providers often provision a little above the advertised tier, so use your own numbers.
The starting rule:
- Download: as of November 2026, a common starting point is about 85 to 90% of measured. Cable needs more headroom here because the provider's CMTS queue sits beyond your router.
- Upload: as of November 2026, a common starting point is about 90 to 95% of measured.
Say you measure 400 Mbps down and 20 Mbps up. At 90%, the download limit is 360 Mbps, which luci-app-sqm and most firmware want as 360,000 kbps. At 92%, the upload limit is 18,400 kbps.
As of November 2026, the commonly recommended overhead for DOCSIS providers is 18 bytes, mpu 64, with no ATM compensation. In CAKE that's the single keyword docsis. As of November 2026, with fq_codel or simple.qos, the commonly recommended setting is the Ethernet link layer with 18 bytes of overhead.
Then tune. If the grade is still B or worse, lower the limit on the direction that bloats in 5% steps. If you get A+, try raising it 2 to 3% to win back speed. The two calculators below do the arithmetic.
SQM download limit for your cable line
Example: Measured download speed: 400 Mbps, Share of measured speed to allow: 90 %. Result: 360,000 kbps.
Measured Mbps x 1,000 x headroom %. For DOCSIS cable, set overhead 18 and mpu 64, or just add the keyword 'docsis' in CAKE. Cable needs more download headroom because the ISP's CMTS queue sits beyond your router.
Can your rented cable gateway do SQM, or do you need your own router?
The rented all-in-one gateways, the Xfinity xFi Gateway and Cox Panoramic Wifi, have no SQM settings. You need a router you control behind them.
- Xfinity: as of 2026, sign in at 10.0.0.1 and look for Bridge Mode under Gateway > At a Glance, or use the Xfinity app; labels vary by firmware. Plug your router into the gateway. Bridge mode kills the gateway's Wi-Fi, which is the point.
- Cox Panoramic: enable bridge mode in the gateway's admin page, or ask Cox support to do it.
- Spectrum: most installs use a separate modem plus a Spectrum Wi-Fi router. Keep the modem, swap the router. No bridge mode needed.
The other route is buying your own DOCSIS 3.1 modem from your provider's approved list. That guarantees PIE on the upstream, and on Xfinity and Cox it can end the gateway rental fee.
Where is the SQM setting on your router?
The table below gives the exact path and the setting to pick. The short version: OpenWrt and GL.iNet use luci-app-sqm, UniFi calls it Smart Queues, eero tucks it into eero Labs as "Optimize for Conferencing and Gaming," Asuswrt-Merlin offers CAKE, Firewalla has Smart Queue, and pfSense and OPNsense use FQ_CoDel limiters. Stock Netgear firmware has nothing, though DumaOS Nighthawk Pro Gaming models have Anti-Bufferbloat.
eero's automatic mode picks its own limits. Easy, but you can't fine-tune the upload, the side that matters on cable.
As of 2026, Smart Queues on older UniFi gateways such as the USG are reported to limit throughput well below gigabit. Check your model's limit before switching it on for a fast plan.
Which routers can fix bufferbloat, and where the setting is
| Router or firmware | SQM type | Menu path | Setting to pick |
|---|---|---|---|
| OpenWrt (any supported router) | CAKE or fq_codel (luci-app-sqm) | Network > SQM QoS (install luci-app-sqm under System > Software) | cake + piece_of_cake.qos, WAN interface, Ethernet overhead 18 or 'docsis' |
| GL.iNet | CAKE or fq_codel (OpenWrt base) | Advanced Settings (LuCI) > Network > SQM QoS | Same as OpenWrt |
| Ubiquiti UniFi gateways | fq_codel (Smart Queues) | Settings > Internet > your WAN > Smart Queues | On, enter download and upload limits; check the model's throughput cap |
| eero | Automatic SQM | eero app > Settings > eero Labs > Optimize for Conferencing and Gaming | On (limits set automatically) |
| ASUS with Asuswrt-Merlin | CAKE | Adaptive QoS > QoS > QoS type: Cake | Manual bandwidth, overhead preset DOCSIS |
| ASUS stock firmware | App priority (Adaptive QoS); not true SQM | Adaptive QoS > QoS | Traditional QoS with manual limits, or flash Merlin |
| Netgear Nighthawk (stock) | No SQM (Dynamic QoS ranks apps) | Advanced > Setup > QoS | Use your own SQM router instead |
| Netgear Nighthawk Pro Gaming (DumaOS) | Anti-Bufferbloat | QoS > Anti-Bufferbloat | Always, with limits below measured speed |
| Firewalla | fq_codel or CAKE (Smart Queue) | Firewalla app > Network > WAN > Smart Queue | On, set limits from the calculator |
| pfSense / OPNsense | FQ_CoDel limiters or shaper | pfSense: Firewall > Traffic Shaper > Limiters; OPNsense: Firewall > Shaper | FQ_CODEL scheduler, limits from the calculator |
| Xfinity xFi Gateway / Cox Panoramic | - | Admin page or app > Bridge Mode | Bridge mode, then your own SQM router behind it |
| Spectrum Wi-Fi router | - | Replace it; keep the Spectrum modem | Your own SQM router plugged into the modem |
Menu paths from each vendor's firmware and documentation as of 2026; labels shift between firmware versions.
How do you turn on SQM in OpenWrt and prove it worked?
- Run the Waveform bufferbloat test on a wired computer. Write down the grade and the download and upload latency.
- Go to System > Software, update the lists and install luci-app-sqm.
- Open Network > SQM QoS and enable it on the WAN interface, the port facing the modem.
- Enter the download and upload limits in kbps from the calculators.
- In the Queue Discipline tab, pick cake with piece_of_cake.qos, or fq_codel with simple.qos on a weak CPU.
- As of November 2026, in Link Layer Adaptation, a commonly recommended choice is Ethernet with 18 bytes of overhead, or the
docsiskeyword in CAKE. - Save & Apply.
Rerun the same test at the same hour. Success is an A or A+, with loaded latency within a few tens of milliseconds of idle. Then recreate the case that was actually broken: start a big cloud backup and join a game or a video call at the same time.
The mistakes we see most:
- shaping the LAN interface instead of WAN
- typing Mbps where kbps is expected (20 instead of 20,000 caps you below dial-up)
- swapping download and upload
- testing over Wi-Fi, where a Wi-Fi bottleneck hides the result
Watch the CPU during the test. If one core pins at 100% and throughput collapses, switch to fq_codel or lower the limits.
The test, shape, retest loop
- Baseline testWired bufferbloat test, note grade and loaded latency
- Set limitsDownload 85 to 90%, upload 90 to 95% of measured, in kbps
- Enable CAKE with docsisWAN interface, overhead 18, mpu 64
- Retest under upload loadSame test, same hour, with a big backup running
- Tune in 5% stepsLower the side that bloats; raise 2 to 3% if you hit A+
Does Xfinity's low latency DOCSIS (L4S) make SQM unnecessary?
No. As of 2025, Comcast was reported to be rolling out low latency DOCSIS with L4S on Xfinity after trials in several markets. Traffic that marks itself as L4S gets a separate short queue through the modem and the network. As of 2025, Comcast has named several application partners for its low latency program.
The catch: only apps and devices that support L4S get the short lane. Most games, Discord, Zoom and the backup clogging your upload still ride the classic queue. A saturated home router or weak Wi-Fi sits outside what L4S touches. Spectrum and Cox customers mostly can't count on it yet.
Read next: Home
Frequently asked questions
Does SQM help if the lag is on Wi-Fi?
Partly. SQM fixes the queue on the link to your provider. A weak or crowded Wi-Fi signal adds delay of its own, so test wired first. If wired grades A and Wi-Fi doesn't, move the router or add a wired access point.
Should SQM stay on all the time?
Yes. The cap matters most when the line is near full, which is exactly when your calls and games suffer. As of 2026, your speed test peak will typically read an estimated 5 to 15% lower. Your ping under load won't climb.
Do I need new limits after a plan change or a network upgrade?
Yes. Re-measure and run the calculators again whenever your plan changes or your provider raises upload speeds in your area, then retest. Limits set for an old 35 Mbps upload will waste most of a faster one.