Myth vs Fact: The 23 Percent of Network Traffic Nobody Budgets For
Archi4 min read
Contents
- The Number That Keeps Showing Up in Audits
- Myth One: Video Is Always the Bandwidth Hog
- What the flow data shows
- Why the myth persists
- Myth Two: Encrypted Traffic Is Opaque, So Give Up
- Myth Three: Capacity Planning Means Buying More Bandwidth
- The scheduling question
- Myth Four: The Unknown Bucket Is Small Enough to Ignore
- What to Do With the Number
- Practical steps that cost little
- What good looks like
The Number That Keeps Showing Up in Audits#
Ask a network engineer what consumes the most bandwidth and the answer is usually video. Streaming, conferencing, surveillance uploads. Reasonable guess. Then the traffic analysis comes back and a stubborn figure appears: roughly 20 to 25 percent of traffic on a typical enterprise or carrier link is unidentified or misclassified. Call it 23 percent as a working midpoint. It is shadow traffic, and almost nobody plans capacity for it.
That gap is not a measurement glitch. It is the sum of protocols that evade port-based classifiers, applications that tunnel over HTTPS, backup jobs scheduled by someone who left two years ago, and IoT devices phoning home on odd schedules. Each piece looks small. Together they form a block of traffic large enough to distort every forecast built on top of it.
Myth One: Video Is Always the Bandwidth Hog#
What the flow data shows#
Video does matter, especially in offices with heavy conferencing. But aggregate flow records from a mixed environment tend to show a long tail that rivals it. Think about what lives on a corporate link: endpoint telemetry, software update distribution, DNS lookups, certificate revocation checks, cloud sync clients, guest devices, printers, badge readers, HVAC controllers. None of these are headline applications. All of them generate constant background chatter.
The 23 percent figure is an aggregate, not a single culprit. It is the long tail behaving like a single tenant.
Why the myth persists#
Dashboards report what they can classify. Video traffic is easy to identify because it uses recognizable protocols and steady bitrates. Shadow traffic is by definition hard to label, so it lands in an "unknown" bucket that many teams quietly ignore. The chart looks clean because the messy part has been pushed off the page.
Myth Two: Encrypted Traffic Is Opaque, So Give Up#
Encryption does block payload inspection, but metadata remains. Flow records still reveal source, destination, packet size distribution, timing, and TLS handshake details. The Server Name Indication field, present in most handshakes, often identifies the destination service even when the payload is sealed.
A backup agent pushing nightly to an object storage endpoint looks very different from a video call, even if both are encrypted. Packet size patterns alone separate bulk transfer from interactive sessions. Teams that treat encryption as a dead end are leaving the most useful evidence on the table.
Myth Three: Capacity Planning Means Buying More Bandwidth#
Adding a bigger pipe is the default response to congestion. It is also the most expensive way to solve a problem that is often a scheduling issue. If a large fraction of your traffic is unmanaged, you are not buying capacity for applications. You are buying headroom for things you cannot name.
The scheduling question#
Ask when the shadow traffic runs. Backups, update pushes, and sync jobs often cluster in the same overnight window. Consolidation storage replication may overlap with them. A link that sits at 40 percent average utilization can still spike to saturation because three unmanaged jobs collided at 2 a.m.
Shifting one job by four hours costs nothing. Upgrading a circuit costs a budget cycle.
Myth Four: The Unknown Bucket Is Small Enough to Ignore#
Twenty-three percent is not a rounding error. On a 10 Gbps uplink it is roughly 2.3 Gbps of traffic nobody has accounted for. If that traffic is unmanaged, it is also unprioritized. When congestion hits, it competes on equal terms with voice and video, and it may well win because it was not shaped.
There is a security angle too. Unidentified traffic includes devices nobody inventoried and software nobody approved. A printer with a default password, a smart sensor with a hardcoded update server, a forgotten test VM: these are the residents of the unknown bucket.
What to Do With the Number#
Treat 23 percent as a starting hypothesis, not a universal constant. Your own figure will differ, and finding it is the exercise.
Practical steps that cost little#
, Compare byte counts from interface counters against the sum of bytes attributed by your flow collector. The difference is your measurement gap, and it is often larger than expected., Sample full packets, not just headers, for a short window. Even a one-hour capture at a busy aggregation point reveals recurring destinations., Read the SNI field in TLS handshakes. Most collectors export it or can be configured to., Ask the backup and update teams for their schedules and target throughputs. Cross-check those numbers against observed overnight traffic.
What good looks like#
A healthy classification rate is not 100 percent. Some traffic will always resist labeling. But a named, scheduled, and shaped unknown bucket is a very different thing from an ignored one. The goal is to shrink the bucket until it matches what you actually know about your network, not to make the dashboard look complete.
The surprise is not that a fifth of your traffic is unaccounted for. The surprise is how long it has been that way while the capacity conversation kept circling video.
More articles