Traffic classification for operators and ISPs
Network traffic classification engine for telecom operators
Network traffic classification assigns each flow on a network to the application or protocol it belongs to, so that capacity, policy and regulatory decisions rest on what the traffic is, not only on how much of it there is. On operator networks, a port number says little about the application, because many applications share the same common ports, and encryption hides the payload patterns that older engines matched. Interkey Intelligent DPI classifies flows from their behaviour with an AI traffic model built into the engine core, including encrypted and tunnelled flows, without decrypting them.
A Saudi product, built by Interkey
For network engineers, traffic analysts and policy teams at telecom operators, ISPs and regulators. Reviewed November 2026.
- What it answers:which application or protocol a flow belongs to, how much traffic each class carries, and which policies a flow triggered.
- How:from flow behaviour (handshake structure, packet sizing and timing, and endpoint behaviour), not from port numbers or payload alone.
- Encrypted traffic:classified without being decrypted.
- Where it runs:on the operator's own infrastructure; out-of-band to classify and report, inline when policy must act on traffic.
What a traffic classification engine does
A classification engine looks at the flows passing the points of the network it covers and assigns each flow to an application or protocol class. Everything downstream depends on that label: a capacity report that says which applications fill the busy hour, a policy that applies to one class of traffic and not another, and a regulatory report that describes traffic in the categories an obligation requires.
| Step | In Interkey Intelligent DPI |
|---|---|
| Input | Traffic at the network points covered by the deployment, on fixed and mobile networks |
| Classification | Each flow is assigned to an application or protocol class by the AI traffic model |
| Output | Metadata: the application class, traffic volumes and the policies a flow triggered |
| Who uses it | Capacity planning, policy owners, regulatory reporting and network operations teams |
How long that metadata is kept, and who can query it, are set in the deployment design and the contract, following the operator's data-governance rules.
Classification methods compared
Classification engines have used four broad methods. What has changed on operator networks is how much traffic each method can identify on its own.
| Method | How it identifies traffic | Where it falls short |
|---|---|---|
| Port-based | Maps the transport port to an application | Many applications share the same common ports, so the port says little about the application |
| Payload patterns | Matches recognisable patterns in packet content | Encryption hides the payload, so the pattern cannot be seen |
| Flow behaviour | Uses how a flow behaves: how the session is set up, packet sizes and timing, and the services it connects to | Needs several features combined, and must stay current as applications change |
| AI model on flow features | Learns application classes from combinations of behavioural features | Results depend on the traffic the model meets, so they must be measured on the network where it runs |
Interkey Intelligent DPI classifies from flow behaviour, not from port numbers or payload alone, and combines the features in an AI model.
How Interkey Intelligent DPI classifies traffic
- An AI model in the engine core.The AI traffic large model is built into the DPI engine, not added on top of it.
- Observable characteristics.Each flow is classified from characteristics that stay visible even when traffic is encrypted: handshake structure, packet sizing and timing, and endpoint behaviour.
- Multi-modal features, learned automatically.Automated online probing extracts multi-modal features from live traffic, and the model is retrained automatically on them, so it can follow applications that change their behaviour.
- A closed loop.Results shown in Live View flow back to refine models and policies.
- Carrier scale.The engine is designed for traffic intake at terabit-per-second scale across fixed and mobile networks.
Model and policy updates should enter a live network through the operator's change control, and accuracy should be re-checked after significant updates. For the encrypted and evasive cases in more depth, see Encrypted traffic classification without decryption.
What operators use classification for
| Governance job | What classification adds |
|---|---|
| Congestion and capacity planning | Which applications drive load, and how that changes over time. See Mobile network congestion: diagnose it by application |
| Policy enforcement | Rules that apply per application, subscriber group or cell, in support of plans, fair-use terms and QoS rules |
| Content governance for ISPs | Application classification and content filtering to meet national regulatory requirements |
| Regulatory reporting | A reporting layer for capacity and regulatory teams, with traffic described by application |
| Security visibility | Tunnelled traffic, and traffic that does not behave like the application it presents itself as |
Out-of-band to classify, inline to act
Classification and reporting work out-of-band, where the platform observes traffic without sitting in the forwarding path. A policy acting on traffic needs an inline deployment, which operators usually add as a second phase, after classification has been validated on their own traffic. See Inline vs out-of-band DPI: choosing the deployment.
The platform runs on the operator's infrastructure and can operate entirely in the Kingdom. Interkey Intelligent DPI is Interkey's DPI product; Interkey's team in Riyadh handles localisation, integration and support, in Arabic and English.
Five questions to ask of any classification engine
1. Which classes matter to you?List the application classes your policies and reports depend on before you compare engines.
2. How is accuracy reported?Ask for detection and false-positive rates for each class, measured on traffic you have verified, not a single headline figure. See Measuring DPI classification accuracy on your traffic.
3. How does it keep up?Ask how the model learns applications that change, and how updates enter your change control.
4. What leaves the platform?Confirm that the output is metadata, and that retention and access follow your rules.
5. Where does it run?Confirm that it runs on your infrastructure, and where traffic and results are kept.
To turn these questions into tender wording, see the DPI RFP requirements checklist.
Questions about network traffic classification
What is network traffic classification?
Is traffic classification the same as deep packet inspection?
Can encrypted traffic be classified?
What does the engine output?
How is it priced?
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.
Discuss classification on your network
Tell us which application classes matter for your policies and reports, where in the network you need visibility (RAN, core or peering), 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