Bufferbloat

What Bufferbloat Is

Bufferbloat is the excessive delay that appears on a network connection when a router, modem, or other device queues too much data in its buffers instead of dropping or pacing packets appropriately.

It was formally identified and named around 2010 by Jim Gettys and Van Jacobson, who traced mysterious latency spikes on home networks to oversized buffers in consumer hardware, buffers that were originally added with good intentions, to prevent packet loss, but that instead let queues balloon uncontrolled during periods of high traffic.

\n\nThe practical effect shows up whenever a connection is pushed to its limit.

A single large upload, like backing up a phone to iCloud or seeding a torrent, can fill a router's send buffer so completely that every other packet, including a Zoom call or a Call of Duty match, has to wait in line behind it.

Round-trip latency that normally sits at 10 to 20 milliseconds on a Verizon Fios or Comcast Xfinity connection can spike to several hundred milliseconds or more, even though a raw speed test still reports the full subscribed bandwidth.

This is why a household can pay for a gigabit plan from Google Fiber or a 500 Mbps cable tier from Spectrum and still experience choppy video calls and laggy online gaming the moment someone else starts a large download.

• Router queue depthConsumer routers like the TP-Link Archer or Netgear Nighthawk series often ship with WAN buffers sized for 100 Mbps-era links, so on a 500 Mbps fiber plan they hold far more data than needed, adding latency.
• Upload-heavy triggersSaturating the upload path, such as backing up photos to Google Photos or pushing a large file to Dropbox, fills the buffer fastest since upload bandwidth on cable and DSL lines is typically much lower than download.
• ISP modem contributionCable modems from Comcast Xfinity or Spectrum and DSL gateways from AT&T or CenturyLink add their own buffering layer on top of the router, compounding delay under load.
• Application symptomsZoom and Microsoft Teams calls break up, Call of Duty or Valorant players see ping spike from 20ms to 400ms, and Alexa or Google Home commands lag, all while a download bar keeps climbing normally.
• Measurement gapStandard speed tests from providers like Ookla historically reported only throughput, missing bufferbloat entirely, which is why latency-under-load metrics matter.

A practical way to grasp bufferbloat is to run a large upload, such as syncing a folder to OneDrive, while simultaneously pinging a server; if idle latency is 15ms and it jumps to 600ms under that load, the connection has significant bufferbloat, even though a standard speed test would still report full advertised throughput and show nothing wrong.

Why Bufferbloat Happens

Bufferbloat happens because network hardware, from cable modems to Wi-Fi routers, is built to avoid dropping packets at almost any cost.

Every router, modem, and switch along your path holds a small memory queue, or buffer, to absorb short bursts of traffic so a sudden spike does not cause instant packet loss.

That is a reasonable design goal, but many devices, including budget routers from vendors like TP-Link and Netgear and cable modems from Arris, ship with buffers sized for outdated assumptions about network speed.

When those buffers are far larger than they need to be, they stop absorbing brief bursts and start queuing sustained traffic for seconds at a time.

The deeper issue is how internet protocols were designed to respond to congestion. TCP, the protocol behind most web browsing, file transfers, and video calls, relies on packet loss as its primary signal to slow down.

When buffers are oversized, packets do not get dropped, they just wait in line. TCP senses no problem and keeps transmitting at full speed, filling the buffer further.

The result is latency that climbs from 20 milliseconds to 500 milliseconds or more during a large upload, even though no packets are technically lost.

On Internet Speed Test, this shows up as a jitter and latency spike specifically during throughput-heavy test phases, a telltale bufferbloat signature you will not see in a standalone speed test that only measures peak megabits per second.

1. Oversized buffers by designChipset vendors like Broadcom and Qualcomm Atheros, along with router makers such as TP-Link, Netgear, and Asus, historically built network interfaces with large first-in-first-out queues to prevent packet loss during brief traffic bursts. Memory got cheaper, so buffer sizes grew, but the queue management logic never kept pace.
2. The upload bottleneck at the modemCable modems from Arris and Motorola, and DSL gateways from providers like CenturyLink and Frontier, often carry the deepest queues on the slowest link in the path, the last-mile connection. When you upload a large file on Zoom or push a Dropbox sync, that modem buffer fills before the ISP's edge router even notices.
3. TCP's blind trust in packet loss signalsStandard TCP congestion control, including Reno and Cubic, only slows down when it detects dropped packets. Deep buffers delay that signal for seconds, so senders keep pumping data into an already saturated link, unaware they are the cause of the problem.
4. Router firmware without active queue managementMany consumer routers, including older Linksys and D-Link models, ship without AQM algorithms like CoDel or PIE enabled by default. Without something actively dropping or marking packets before the queue overflows, bufferbloat runs unchecked.
5. Wi-Fi and mesh systems adding their own queuesMesh systems like Google Nest Wifi, Eero, and Orbi introduce additional buffering at each hop to manage contention on shared radio spectrum, so bufferbloat can compound across a multi-node home network even when the internet connection itself is healthy.

Bufferbloat is ultimately a mismatch between hardware built to be forgiving and software congestion control built to assume queues are short.

A household running an Asus router with a saturated 20 Mbps DSL line from CenturyLink can see ping times jump from 15 ms to over 800 ms the moment someone starts a large OneDrive backup, even though the download speed test itself still reports numbers close to the advertised plan.

Fixing this requires active queue management like CoDel or FQ-CoDel, now built into OpenWrt and some newer firmware from Ubiquiti and Netgear's Orbi line, which deliberately manages queue depth instead of letting it grow unchecked.

How to Tell If You Have Bufferbloat

Bufferbloat rarely announces itself with a dramatic speed drop, which is why so many people misdiagnose it as a bandwidth problem and pay for a bigger plan that doesn't fix anything.

The real signal is a specific pattern: your download and upload numbers on a speed test still look fine, maybe 300 Mbps on a Comcast Xfinity or Virgin Media connection, but the moment that test starts pushing data, everything else on the network stutters.

Zoom calls freeze mid-sentence, Discord voice cuts out, and pings to a game server spike from 20ms to 400ms.

The clearest tell is comparing idle latency to loaded latency. If a ping to 1. 1. 1. 1 or 8. 8. 8.

8 sits around 15-25ms normally but jumps past 200-300ms the instant a large download or a speed test starts, your router's buffer is filling up and queuing packets instead of dropping or prioritizing them.

This happens most often on older cable modems, budget routers running stock firmware, and connections where the ISP-supplied gateway, like an Xfinity xFi or a basic Sky Hub, lacks any active queue management.

1. Step 1Run a bufferbloat-specific test rather than a plain speed test. Waveform's Bufferbloat Test, the DSLReports speed test, or Internet Speed Test's latency-under-load results all measure ping response while simultaneously saturating your upload and download paths.
2. Step 2Check the grade, not just the number. Waveform and DSLReports assign A through F grades based on how much latency increases under load. Anything below a C, meaning latency spikes over 100-200ms during the test, points to bufferbloat.
3. Step 3Isolate the cause by testing with one device wired directly into the modem via Ethernet, bypassing Wi-Fi entirely. If the problem disappears, weak Wi-Fi signal or an overloaded 2.4GHz band on an older TP-Link or Netgear router is compounding the issue.
4. Step 4Repeat the test at different times of day and on both wired and wireless connections. Consistent spikes during evening peak hours alongside a fixed baseline suggest local queue buildup rather than ISP-side congestion.
5. Step 5Cross-check with a real-world symptom: start a large download or an Xbox Series X game update, then immediately try a Zoom call or online match in Valorant. Audible stutter or rubber-banding confirms what the test numbers already showed.

If your Waveform or DSLReports grade comes back D or F, the fix usually isn't a faster plan but Smart Queue Management, available natively in OpenWrt-based firmware, on Ubiquiti's UniFi Dream Machine, or through routers that support fq_codel and cake, which actively manage buffer size instead of letting it grow unchecked, typically dropping loaded latency from 300ms back down under 50ms without touching your subscribed download speed.

Bufferbloat and Remote Work

Remote work depends on a steady, low-latency connection far more than it depends on raw bandwidth, and this is exactly where bufferbloat does the most damage.

A worker on a 300 Mbps fiber plan from Verizon Fios or AT&T can still suffer choppy Zoom calls if their router's buffers fill up during a routine background task like a Windows update or a Dropbox sync, because those oversized buffers delay every other packet waiting in line, including the voice and video data a call depends on.

This matters most during the exact moments remote work demands reliability: a client call, a screen share during a sprint review, or a VPN session into a corporate network.

Standard speed tests report peak throughput and miss the problem entirely, which is why bufferbloat has become a common but frequently misdiagnosed cause of "my internet is fast but my calls still freeze" complaints among people working from home on connections from providers like Comcast Xfinity, Spectrum, or Vodafone.

• Video call freezes on sendUploading a large file in the background, say syncing a folder to OneDrive or Google Drive, fills the router queue and delays outgoing Zoom or Microsoft Teams packets, causing your own video to freeze for the other participants even though your download speed test shows no problem.
• Screen sharing lags behind audioWhen bufferbloat hits the upstream link, screen share frames queue up behind other traffic, so colleagues see your cursor move seconds after you narrate it, a classic symptom on cable connections from Comcast Xfinity or Spectrum during peak evening hours.
• VoIP calls cut out mid-sentenceCisco Webex and RingCentral calls rely on steady low latency, and a bloated buffer on a budget router like a TP-Link Archer or older Netgear Nighthawk can spike latency from 20ms to over 500ms the moment someone else on the network starts a large download.
• VPN sessions time outRemote workers connecting through Cisco AnyConnect or a corporate WireGuard VPN often see disconnects under load, because bufferbloat-induced latency spikes exceed the VPN's keepalive tolerance, forcing a reconnect and dropping an active session.
• Cloud collaboration tools stallReal-time editing in Google Docs or file syncing in Dropbox can appear to hang for several seconds during a bufferbloat episode, since small status packets get stuck behind bulk data in the same queue.
• Latency-sensitive apps misfireSlack huddles and Teams calls that use adaptive bitrate will downgrade video quality automatically when jitter rises, even if your measured bandwidth is more than sufficient for high definition video.

The fix for most home offices is straightforward: enable Smart Queue Management or an equivalent active queue management feature, available on OpenWrt-based routers, GL.

iNet travel routers, and newer firmware from ASUS and Netgear, or switch to a router that supports SQM out of the box.

Running a bufferbloat-specific test, such as the one built into Internet Speed Test alongside standard download and upload results, before and after enabling SQM shows the difference immediately, often cutting latency under load from several hundred milliseconds to under 30ms, which is the threshold where video calls stop stuttering.

Smart Queue Management and QoS Explained

Smart Queue Management, often abbreviated SQM, is the modern fix for bufferbloat.

Where old-style QoS simply prioritized certain traffic types, SQM actively manages queue depth on both the upload and download side of a connection, using algorithms like CAKE or fq_codel to keep latency low even when a link is fully saturated.

Firmware projects such as OpenWrt and pfSense build these algorithms in natively, and mesh systems like eero and TP-Link Deco have started shipping simplified versions under names like "bufferbloat protection" or "traffic prioritization. "

The core idea is setting a shaped rate slightly below your ISP's advertised speed, typically 90 to 95 percent, so the router's queue becomes the bottleneck instead of the ISP's modem buffer.

A Comcast Xfinity or Cox gigabit connection, for example, often has a deep buffer in the cable modem itself; without shaping, that buffer fills during large downloads and adds hundreds of milliseconds of delay to every other packet.

SQM intercepts that congestion earlier, at a point where it can be managed intelligently rather than left to overflow.

• CAKE (Common Applications Kept Enhanced)The most widely recommended algorithm, built into OpenWrt and GL.iNet routers, combining rate shaping, fair queuing, and DSCP-aware prioritization in one package.
• fq_codelA lighter-weight predecessor to CAKE, still used in many consumer routers and Linux kernels, that separates flows and drops packets proactively to prevent queue buildup.
• DSCP tagging and traffic classesVoice and video calls on Zoom, Microsoft Teams, or Discord can be tagged for priority handling so a large Steam or Xbox game download does not stall a work call.
• Per-device bandwidth allocationEnterprise-grade gear from Ubiquiti UniFi or Netgear Orbi Pro lets administrators reserve minimum bandwidth per device, useful in homes with multiple remote workers.
• Active Queue Management on ISP hardwareSome providers, including certain Deutsche Telekom and BT Openreach deployments, now enable AQM on the CPE itself, reducing the need for router-side configuration.

Configuring SQM correctly requires knowing your true sustained throughput, not just the number printed on your ISP bill, which is exactly what a real-time test on Internet Speed Test provides.

Run the test during peak hours, note the actual download and upload figures alongside jitter, then set your router's shaper to roughly 95 percent of those measured values. Enthusiasts running OpenWrt on a GL.

iNet Flint 2 or a Netgear Nighthawk flashed with third-party firmware routinely report ping spikes dropping from 400 milliseconds to under 30 milliseconds under load once CAKE is properly tuned to actual, tested line speed rather than marketing numbers.

How to Fix Bufferbloat

Fixing bufferbloat means giving your router direct control over its queues instead of letting an oversized buffer decide the order packets leave your network. The core tool is Smart Queue Management, built on algorithms like fq_codel and cake, both developed out of the Bufferbloat.

net project led by engineers including Dave Taht and Jim Gettys.

These algorithms actively manage queue depth, dropping or reordering packets intelligently so latency-sensitive traffic, such as a Zoom call or an online match in Valorant, does not sit behind a large file download queued on your modem.

\n\nMost consumer routers do not enable this by default, so the fix is a configuration change rather than a purchase, though some hardware makes it far easier than others.

Firmware such as OpenWrt, DD-WRT, and pfSense expose SQM controls directly, while consumer mesh systems from Eero, Google Nest Wifi, or Netgear Orbi often lack granular queue management entirely, which is worth checking before you buy.

1. Enable Smart Queue ManagementTurn on SQM on your router, using fq_codel or cake as the queuing discipline. OpenWrt, pfSense, and firmware like Asus Merlin all expose this under QoS or Traffic Shaping settings.
2. Set bandwidth limits below your sync rateCap upload and download to roughly 85 to 95 percent of your measured throughput. Run a Internet Speed Test test first, then enter those numbers into your router's SQM fields so the router controls the queue instead of your ISP's modem.
3. Replace ISP-supplied combo modems where possibleDevices like the Arris SURFboard or Xfinity xFi Gateway often use large, unmanaged buffers. Pairing a separate modem with a router that supports cake, such as a GL.iNet Flint 2 or a Ubiquiti Dream Machine running third-party firmware, gives you more control.
4. Update router firmwareManufacturers including Netgear, TP-Link, and ASUS have shipped firmware updates that improve buffer handling and add native QoS features, so check for updates before troubleshooting further.
5. Test under loadUse Internet Speed Test's speed test while running a video call or large download to check latency under load, not just idle ping. A well-tuned connection should hold latency increases under 20 to 30 milliseconds even during saturation.
6. Consider your ISP tier and equipmentProviders like Comcast Xfinity, Spectrum, and AT&T Fiber now ship gateways with built-in Active Queue Management on newer models, so upgrading rental equipment or asking for a firmware refresh can resolve bufferbloat without buying new hardware.

The single most effective fix remains manually enabling cake-based SQM on a router that supports it and setting the bandwidth ceiling slightly under your actual measured speed. Run a baseline test on Internet Speed Test, apply the settings, then retest under load.

If latency under load drops from several hundred milliseconds to under 30, the fix is working. If your current gateway offers no queue management options at all, that alone is a strong reason to replace ISP rental equipment with a dedicated router.

Frequently Asked Questions