What Is Bufferbloat and Why Your Internet Feels Slow Even When It's "Fast"

You upgraded your internet connection. The speed test shows 50Mbps — well above what it was before. And yet the video calls still stutter. The VoIP calls still sound like the other person is talking from underwater. The cloud applications still feel sluggish during the middle of the day.

The speed test says the internet is fast. Your experience says otherwise. Both are correct.

The problem is almost certainly bufferbloat — one of the most common and most misunderstood causes of poor internet performance, and one that bandwidth alone cannot fix.

What Bufferbloat Is

When data travels across a network, it moves in packets. Those packets pass through multiple devices — routers, switches, modems — and each device has a buffer: a small amount of memory used to hold packets temporarily when the device is processing more traffic than it can forward immediately.

Buffers are necessary. Without them, packets would be dropped whenever a device was momentarily busy, causing constant retransmissions and poor performance.

The problem is that modern network equipment — particularly consumer and small-business grade routers — tends to have very large buffers. The manufacturers' reasoning is that larger buffers mean fewer dropped packets. In practice, large buffers cause a different problem: when the buffer fills up, packets sit waiting in the queue for a long time before they are forwarded.

This waiting time is called queuing latency. When a buffer is large and full, queuing latency can be hundreds of milliseconds — far more than the actual transmission time across the network.

This is bufferbloat: high latency caused by oversized, poorly managed buffers in network equipment.

Why It Feels the Way It Does

Bufferbloat has a specific and recognisable symptom pattern.

Under light load — a single user browsing the web or checking email — everything feels fine. Latency is low, applications are responsive.

Under moderate to heavy load — multiple users, a large file upload or download in progress, video calls running simultaneously — latency spikes dramatically. Video calls that were fine suddenly stutter and pixelate. VoIP calls become choppy. Applications that were responsive become sluggish.

The connection is not broken. The bandwidth has not disappeared. But the buffer is full, packets are queued, and the latency-sensitive traffic that makes video and voice work in real time is sitting in the queue behind a large file transfer.

The speed test, meanwhile, shows excellent performance — because the speed test fills the buffer and measures how fast data can be pushed through. It does not measure what happens to other traffic while the buffer is full.

The Technical Cause

At a more technical level, TCP — the protocol used by most internet traffic — is designed to probe for available bandwidth by increasing its transmission rate until packets are dropped. When packets are dropped, TCP backs off and reduces its rate.

On a network with large buffers, packets are not dropped when the buffer fills — they are queued. TCP sees no drops and continues increasing its rate. The buffer grows larger. Latency climbs. And since packets are not being dropped, TCP never backs off.

The result is a buffer that is permanently full, permanent high latency, and a connection that feels slow despite having plenty of bandwidth.

This is why adding more bandwidth does not fix bufferbloat. If you double the bandwidth, TCP simply fills the larger buffer faster. The latency remains high under load.

How Active Queue Management Fixes It

The solution is Active Queue Management (AQM) — algorithms deployed on the router that manage the buffer intelligently rather than allowing it to fill.

Instead of simply queuing packets until the buffer is full, AQM algorithms deliberately drop or mark packets before the buffer fills, signalling to TCP that it should back off before queuing latency becomes a problem. By keeping the buffer short, latency stays low even under heavy load.

Two AQM algorithms have become the standard for modern deployments:

fq_codel (Fair Queuing Controlled Delay) — manages the queue by tracking the time each packet spends waiting. When packets wait longer than a target threshold, fq_codel signals congestion. It also implements fair queuing, ensuring that no single flow can monopolise the buffer at the expense of others.

CAKE (Common Applications Kept Enhanced) — a more recent algorithm that builds on fq_codel with additional capabilities: traffic shaping, flow isolation, and more sophisticated fairness management. CAKE is particularly effective on connections where the upload and download speeds are asymmetric — which describes most business internet connections.

Both algorithms are available on MikroTik RouterOS v7 and higher — the platform IJA uses for business and enterprise deployments.

The Real-World Difference

The improvement from properly deployed AQM is not subtle. It is the difference between a connection that feels fast and one that does not — at the same bandwidth.

A business running a 20Mbps DIA connection with CAKE deployed will typically experience:

• Latency that remains low and consistent under load — typically under 20ms where it was previously spiking to 200ms or more
• Video calls that remain stable when other users are downloading or uploading
• VoIP that sounds clear throughout the working day, not just when the network is quiet
• Cloud applications that remain responsive regardless of what else is happening on the connection

The bandwidth has not changed. The experience has changed dramatically.

Why Most Providers Don't Deploy AQM

If AQM makes connections perform so much better, why is it not standard practice?

For ISPs managing large numbers of subscribers, AQM at scale requires more sophisticated infrastructure and careful per-connection configuration. It is easier — and cheaper — to simply provision more bandwidth and let customers upgrade when they complain.

For equipment manufacturers, large buffers mean fewer visible dropped packets and fewer customer complaints about packet loss. The latency problem that results is less visible and harder for customers to diagnose.

For businesses, the result is connections that test well and perform poorly — and a service industry that responds to complaints with "have you tried running a speed test?"

What This Means for Your Business

If your business uses VoIP, video conferencing, or any cloud application — and most businesses do — bufferbloat is almost certainly affecting your experience on a standard broadband or poorly configured business connection.

The questions to ask your provider or IT team are:

• Is AQM deployed on the router managing our connection?
• Which algorithm — CAKE, fq_codel, or similar?
• Has the shaping rate been correctly configured to match our connection speed?

If the answer to the first question is no, or if the question produces a blank look, you have identified a likely cause of poor performance that more bandwidth will not fix.

The IJA Approach

IJA deploys CAKE active queue management on every MikroTik v7+ CPE we manage for business and enterprise clients. It is part of the standard configuration — not an optional extra.

Combined with IJA Verify's continuous latency monitoring, we can measure the before-and-after improvement directly, confirm that AQM is performing as expected, and alert our team if latency patterns suggest a configuration issue.

The goal is not a connection that tests well. The goal is a connection that performs well — for every application, for every user, throughout the business day.

IJA Technologies manages business internet connections with active queue management as standard. If your connection tests well but feels slow, we can help. Talk to us.