Encrypted and tunnelled traffic
Encrypted traffic classification without decryption
Encrypted traffic classification identifies which application an encrypted flow belongs to without reading its content. Interkey Intelligent DPI does this from characteristics that stay observable when the payload is encrypted (handshake structure, packet sizing and timing, and endpoint behaviour), and it does not decrypt traffic to classify it. This page explains how that works, what the result tells an operator, and how to measure accuracy on your own traffic.
A Saudi product, built by Interkey
For core network engineers at telecom operators and ISPs. Reviewed November 2026.
- No decryption:classification relies on handshake structure, packet sizing and timing, and endpoint behaviour.
- The output is a class, not content:which application or protocol a flow belongs to, its volume and the policies it triggered.
- Evasive applications are learned continuously:features are extracted automatically from live traffic and the model is retrained automatically.
- Accuracy depends on your traffic:measure detection and false-positive rates per class on a sample you have verified.
Why port and payload rules stop working
Traditional classification looked at two things: the port a connection used and recognisable patterns in the packet payload. Encryption removes the payload signal, and many applications share the same common ports, so neither signal identifies an application reliably on its own. As the encrypted share of traffic grows, an operator that classifies only by port and payload sees less and less of its own network.
Some applications go further. VPN and other evasive applications keep changing to avoid detection and disguise their traffic as legitimate services. Where an operator's policy, or applicable regulation, requires such applications to be classified or controlled, this is the classification problem that matters.
What remains observable in an encrypted flow
| Characteristic | What it is | Why it helps classification |
|---|---|---|
| Handshake structure | How the encrypted session is set up: the sequence and shape of the opening exchange | Different applications set up sessions in distinguishable ways |
| Packet sizing and timing | The sizes, order and intervals of packets in a flow | Video, messaging, calls, file transfer and tunnels produce different patterns |
| Endpoint behaviour | Which services a flow connects to, and how those connections behave over time | Applications reach characteristic sets of services |
The AI traffic model is built into the DPI engine core rather than added on top of it. It combines these features (multi-modal feature extraction) instead of relying on any single one, and it applies to tunnelled flows as well as encrypted ones.
How the model keeps up with changing applications
- Automated feature extraction.Automated online probing extracts features from live traffic for training.
- Automated model training.The model is retrained automatically on those features.
- Closed-loop feedback.Live View results flow back to refine models and policies.
Whichever way updates are delivered, they should enter a live network through the operator's change control, and accuracy should be re-checked after significant updates.
What encrypted traffic classification tells you, and what it does not
| Classification tells you | Classification does not |
|---|---|
| Which application or protocol class a flow belongs to | Decrypt traffic or read encrypted content |
| Traffic volumes by application | Decide what is permitted: that is set by your policy and applicable regulation |
| Which policies a flow triggered | Set how long data is kept: retention and access follow your data-governance rules and the contract |
The platform runs on your own infrastructure and can operate entirely in the Kingdom. It is built and offered for network governance and regulatory compliance, not for surveillance of individuals: it identifies classes of traffic and leaves the content encrypted.
What operators do with the result
| Use | How classification helps |
|---|---|
| Capacity planning | Shows which applications drive load, even when most of it is encrypted |
| Policy | Lets policies act per application, subscriber group or cell, on traffic that would otherwise be unidentified |
| Regulatory reporting | Gives regulatory teams a description of traffic by application class |
| Content governance | Supports content filtering to meet national regulatory requirements |
How to measure accuracy on your traffic
Classifier results depend on subscriber mix, application mix, the share of encrypted traffic and how recently the classifier was updated. A classifier tuned on one operator's traffic will not behave identically on another's, so measure it on yours.
1. Build a verified sample.Use traffic where you know the right answer, for example generated by known applications on devices you control.
2. Classify it out-of-band.The platform receives a copy of traffic without touching the forwarding path.
3. Report per class, not one figure.Measure detection and false-positive rates for each application class. A false positive matters as much as a miss: it means a policy acts on the wrong traffic.
4. Repeat after updates.Check that results hold as applications change.
Interkey's product material states 99% recognition accuracy, including for VPN and other adversarial encrypted applications. The figure has not been independently benchmarked and is not a specification for any particular network, which is why we recommend measuring it on your traffic.
From classification test to decision: 3 stages
1. Written acceptance criteria.Agree what is tested, on which traffic, how it is measured and what counts as a pass.
2. Out-of-band classification against verified traffic.The measurement steps above, with false positives measured as carefully as detection.
3. Report-only policy, then the inline decision.Policies show what they would have done before they act; inline enforcement follows only after classification has proven itself.
See also: How Interkey runs a DPI proof of concept
Questions engineers ask
Does Interkey Intelligent DPI decrypt traffic?
Can it read the content of encrypted traffic?
Does it work on tunnelled traffic and VPNs?
Which encryption protocols does it support?
How does it handle an application that changes its behaviour?
Can we test classification before deciding anything?
Proof of concept in three phases
Written acceptance criteria.Agree which rules are tested, on which traffic and subscriber groups, how results are measured and what counts as a pass.
Out-of-band classification against verified traffic.Measure false positives as carefully as detection for every class the rules depend on.
Report-only policy, then the inline decision.Each rule shows what it would have done; enforcement follows only after classification has proven itself.
Test classification on your traffic
Tell us which application classes matter to you, roughly what share of your traffic is encrypted (if known), and your timeline. Our team in Riyadh replies in Arabic or English.
- Our engineers read your request
- We agree the scope of a technical session
- You see the product on a use case you choose