Your tariff promises that video served from the operator's own cache does not touch the data package. Picture a subscriber who watches such video for an evening and then sees the main quota go down anyway. The network did not know the traffic was local, or it knew but counted it in the wrong place.
This is a note for engineers and product owners at mobile operators (MNO) and virtual operators (MVNO) who build or check such a tariff. The standards give building blocks but no ready feature: they do not define zero-rating, and operators build it from charging keys and rules. This note gives a working scheme: how to recognize local traffic, where to count it and what to test.
Is excluding local and CDN traffic the same as zero-rating?
BEREC (the body of European telecom regulators) defines zero-rating like this: an ISP applies a price of zero to the data traffic of a particular application or class of applications. That data does not count towards any data cap. By this definition the criterion is the application or a class of applications. In our reading, for local traffic the criterion is where the traffic comes from: a cache or a node inside the operator's network.
Netflix and Google publish programs for placing caches inside ISP networks. By the description of Netflix Open Connect, appliances can be embedded inside an ISP network and are provided free to qualifying ISPs. The ISP configures BGP sessions with these appliances to steer traffic to them. Netflix names two parts of its program: embedded appliances and settlement-free interconnection (peering). Google's help page describes Google Global Cache as a way for ISPs to serve certain Google content from within their own networks.
The Netflix page describes how traffic reaches the appliance, not how to count it. In this note, "local traffic" means traffic that the operator serves from caches and nodes inside its own network.
For accounting, the task comes down to three results. The traffic is not deducted from the main bucket, or it is counted in a separate bucket, or its statistics go to billing under a special rule. A PCEF (Policy and Charging Enforcement Function, the element that applies charging rules to traffic) can apply such a separate policy to a selected service's traffic.
How does the network know that a flow is local?
There are three ways: by address and subnet, by service signature, and by domain. Each one matches something different, and each has a weak spot. In our view, the weak spots in the table are our own reasoning, not measurements.
First, what the standard gives. By 3GPP, a PCC (Policy and Charging Control) rule contains either a list of flow filters or an application identifier, and a charging key when the rule charges. Detection of the application is another matter. In TS 23.203 that logic is not standardized. It is entirely internal to a TDF (Traffic Detection Function) or to a PCEF enhanced with ADC (Application Detection and Control). Put simply, the standard does not define the detection logic itself; it is left to the implementation. Since Release 14, TS 23.203 does standardize how detection inputs, packet flow descriptions such as a server 3-tuple, a URL or a domain name, are provided to the PCEF or TDF via the PFDF (Packet Flow Description Function).
| Method | What is matched | What the standard says | Weak spot |
|---|---|---|---|
| Address and subnet | Prefixes of the caches and nodes, as flow filters | A rule holds a list of flow filters; the one zero-rating example in TS 23.203, clause 4.2.3 (EXAMPLE 4), uses a port and address filter from the subnet of the operator's DNS nodes | An address list needs constant updating; every change in cache prefixes must reach the rule. A cache node may serve more than the promised content: Google Global Cache serves “certain Google content”, not only video, so an address match zero-rates all of it |
| Service by signature | The type of traffic, recognized by the application detection logic | The rule can carry an application identifier; the detection logic is internal to the implementation | A signature tells which service the flow belongs to, not which server delivered it |
| Domain | The domain name against the TLS SNI extension when available, then the certificate's Subject Alternative Name, then the Common Name | Informative Annex X of TS 23.203, described for sponsored services; IP address ranges are checked first | The annex assumes an agreement between the operator and the content provider, and HTTPS with Web PKI; where the name is not visible, for example with TLS Encrypted Client Hello, the method has nothing to match |
Annex X belongs to the standard and describes no particular product. Read its assumptions before you rely on it: if the IP ranges equal "Any", any traffic is said to be compliant. So Annex X does not say that SNI identifies the application in general: it holds only under these assumptions.
Encryption narrows the choice further. A DPI platform cannot read the encrypted payload, so it relies on unencrypted metadata and behavioral and heuristic analysis. For a tariff, the lab has to show what your equipment sees on your own traffic.
In practice the three methods work together. An address list is simple and precise while the list is fresh. A signature does not depend on addresses but names a service, not a server. A domain helps where the name is visible. The mix is the operator's design decision.
Where does the excluded traffic go in the accounting?
Into a bucket with its own key. In 3GPP TS 23.203 a PCC rule can mark a service data flow as not subject to any end user charging, and then it carries no charging key. Rules that do charge share a charging key as the reference to the tariff. On the wire the charging key is carried as Rating-Group on Gx and Gy. RFC 8506 (Diameter Credit-Control) defines a rating group as the identifier that groups services subject to the same rating type.
This gives two designs. In the first, the rule carries no charging key. In the second, a flow in its own rating group is reported on its own. In Diameter credit control, one Multiple-Services-Credit-Control block carries the granted units, the used units and the rating group. From this structure, usage can be reported per rating group on Gy.
Here only the bucket matters. Prices and tariff structure stay with the operator.
What should you check in the lab?
If you build the scenario yourself, we suggest six checks. In our test lab we check charging-exemption scenarios: specific application categories, for example messengers or navigation, must not consume the main data package. The PCEF brochure lists quota exhaustion and charging exceptions among the scenarios to test on the operator's own cases.
- The cache subnet reaches the rule. Send traffic from a cache address and check that the flow matches the exclusion rule and that the main bucket counters stay still.
- The service is recognized and lands in its own bucket. Generate traffic of the service and check that the usage appears under the exclusion's rating group in the Gy report, not under the main one.
- The main bucket is exhausted and the excluded category still works. Run the main bucket down to zero, then repeat checks 1 and 2. Ordinary traffic must follow the tariff's rule for exhausted quota, the excluded category must not.
- A list or signature update does not break the rule. Change the address list or load a new signature set while a session is open, and check that the rule keeps matching and the counters do not move to the main bucket.
- Tethering does not move the flow to another bucket. Repeat check 2 from a device that shares its connection and check that the usage stays under the exclusion’s rating group.
- The exclusion survives an OCS (online charging system) outage. Make the OCS unreachable while the session uses a local quota and check that the excluded traffic still does not consume it.
In our experience, a rule that passes checks 1 to 3 can still fail after an update of the address list or signatures.
How our PCEF handles zero-rating
In our experience, a common mistake in tariffs that zero-rate third-party services, such as messengers or video platforms, is to recognize them by an IP address list instead of signatures. The provider’s addresses change, the list ages, and traffic starts to be charged against the package despite the promise to the customer. For the operator’s own caches it is the other way round: their addresses are stable and under the operator’s control, so an address or AS list is a valid criterion.
Our PCEF recognizes a third-party service such as a messenger or a video platform primarily by traffic signatures in Stingray Service Gateway rather than by an IP address list. Where the criterion is the source, such as the operator’s own caches, the traffic can be defined as a custom protocol by IP, subnet or AS and mapped to its own rating group. It counts that traffic in its own bucket (rating group), and the OCS applies the zero rate to that rating group. Usage in this rating group is reported separately, so when the OCS rates it at zero, it does not reduce the main quota. Cloud protocols in Stingray Service Gateway are updated from VAS Cloud, by default once a day, so they do not wait for a manual list change. Built-in protocols are updated with the software version.
Each traffic category goes into its own bucket with its own limit and unit of measurement. Quota exhaustion is handled per bucket, so an exhausted video package does not affect messengers. Buckets under online charging are synchronized with the OCS via Gy.
Our PCEF brochure lists excluding local CDN traffic from the quota among its zero-rating scenarios.
If you assemble this scenario on your own equipment, start with the six checks and run them on your real prefixes and your real services.
Specifications: 3GPP TS 23.203 and TS 29.212, RFC 8506. Programs and definitions: Netflix Open Connect, Google Global Cache, BEREC: what is zero-rating. More about our PCEF: vasexperts.com/products/pcef-policy-and-charging-enforcement-function.
