Content for  TS 29.500  Word version:  16.4.0

Top   Top   Up   Prev   None
1…   5…   5.2.3   5.2.4…   6…   6.4…   6.5…   6.6…   6.7…   6.10…   6.11…   A…


A  Client-side Adaptive Throttling for Overload Control
This clause contains an example algorithm to make an NF Service Consumer adjust the traffic rate sent to an NF Service Producer based on the number of received "rejects" of HTTP requests with a status code "503 Service Unavailable", or requests that have timed-out and the response was never received. This algorithm is described in the book "Betsy Beyer, et al; Google: Site Reliability Engineering" (, chapter 21, "Handling Overload".
Each client (NF Service Consumer) keeps track of the following counters during a certain time window:
  • Requests: The number of requests that the client (NF Service Consumer) needs to handle. Under normal operation (no overload), all these requests are sent to the server (NF Service Producer). Under an overload situation, part of these requests are locally rejected by the client (and not sent to the server), and the rest of the requests are sent to the server.
  • Accepts: The number of requests accepted by the server (i.e., requests for which a response has been effectively received at the client, with a status code other than "503 Service Unavailable").
    When there is no server overload, these values are equal.
When there is an overload status in the server, the rate between "Accepts" and "Requests" decreases progressively. When this rate falls below a certain point (given by an algorithm parameter named "K"), the client shall start dropping some requests locally and not send them to the server.
The local rejection of requests can be done by calculating a "Client request rejection probability", as:
Equation for Client request rejection probability
So, for example, assuming that the K parameter is set at 1.5:
  • if the server accepts >67% of the traffic, and rejects <33% of the traffic, the client does not take any throttling action, and keeps sending to the server all the traffic it has available for processing
  • if, during a first time-window, the server accepts, e.g., only 60% of the requests, and rejects 40% due to overload, the application of this algorithm implies that the client must drop locally 10% of the requests (probabilistically), and only send to the server the remainder 90% of its traffic.
  • if, during a second time-window, the client keeps the same amount of available traffic to handle, but the server continues rejecting requests with same rate as before (40%) of the received requests, the application of the algorithm again, results in increasing the drop rate to 14.5%, and sending to the server only 85.5% of the available traffic.
The value of the parameter K, along with the size of the time window during which the total number of "requests" and "accepts" is accounted for, has a fundamental role on how the algorithm behaves. If K is higher, the algorithm is more "permissive", and the client does not start dropping requests locally until the rejection rate is higher (e.g., >50%, for K = 2); if K is lower, the algorithm is more "aggressive", and the client starts dropping requests sooner (e.g., K = 1.1 implies to start dropping requests as soon as the server rejects >10% of the requests).
B (Normative)  3gpp-Sbi-Callback TypesWord‑p. 75
This annex specifies the list of allowed 3GPP SBI callback type values for the 3gpp-Sbi-Callback HTTP custom header specified in clause
Table B-1 specifies callbacks that are invoked across PLMN.
Value for 3gpp-Sbi-Callback Custom HTTP Header

3GPP TS 29.502, Clause,, and
3GPP TS 29.518, Clause
3GPP TS 29.525, Clauses  4.2.4, 5.5.2 and 5.5.3.
3GPP TS 29.531, Clause
3GPP TS 29.518, Clause
3GPP TS 29.515, Clause
3GPP TS 32.291, Clause
3GPP TS 29.526, Clause
3GPP TS 29.526, Clause

For notification and callback service operations (used across PLMNs or within a PLMN) that are not part of Table B.1, the value of the header shall be constructed as follows:
"N<NF>_<service name>_<name of the callback service operation in the corresponding OpenAPI specification file>"
Nsmf_PDUSession_smContextStatusNotification (for the Notify SM Context Status service operation)
C  Internal NF Routing of HTTP Requests |R16|
The internal details of the architecture of a Network Function instance is out of the scope of 3GPP and are entirely implementation-specific. This annex describes how an instance of an NF Service Producer can route internally HTTP requests received on a given Service-Based Interface.
Figure C-1 illustrates an example component architecture where incoming HTTP requests are received and processed in a component named as "Ingress Proxy" module and route them to the appropriate computing resource in the NF.
Reproduction of 3GPP TS 29.500, Figure C-1: Internal message routing inside NF Service Producer
The Ingress Proxy may parse any of the different components in the HTTP request, but typically it may parse the path of the URI (i.e. the :path pseudo-header in the HTTP/2 request). Parsing of other component in the request message, such as the HTTP body, is also possible but it is not desirable as it requires the parsing of the entire body (i.e. a JSON document) which is a much more computing-intensive task.
The path component of the URI contains the service name of the requested SBA service, so frequently the routing is done based on this component.
It is also frequent to inspect other components of the path (i.e. path segments), to do a more fine-grained routing and direct requests done on a specific HTTP resource(s) towards a given computing resource(s).
It can be noted that the path components used to determine the target computing resource typically do not need to be statically defined but are frequently defined in terms of "variables", or placeholders, similarly to how they are defined in the OpenAPI specification language (a mechanism usually known as "path templating"). See:
$  Change historyWord‑p. 76

Up   Top