4A SGW/PGW selection using GTP-C load control

29.3033GPPDomain Name System ProceduresRelease 17Stage 3TS

4A.1 General

The nodes responsible for the SGW or PGW selection (i.e. the MME, S4-SGSN, TWAN or ePDG) shall apply the additions specified in this clause, if GTP-C load control is supported and enabled (see clause 12.2 of 3GPP TS 29.274 [23]).

GTP-C load control allows for the advertising of load information at node level (i.e. the SGW or PGW load) or at APN level (for a PGW only) and selecting the target node (i.e. the SGW or PGW) based on this information.

The nodes responsible for the SGW or PGW selection shall take into account the load control information reported via GTP-C signalling and shall implement a SGW or PGW selection logic according to the principles specified in the following clauses.

NOTE: The exact algorithm used at the node performing the SGW or PGW selection, as per the aforementioned requirements, is implementation specific.

4A.2 Node-level load control

This clause specifies the extensions to the SGW or PGW selection procedures which shall be supported when GTP-C load control is supported and enabled at node level.

The node level load information consists of the "Load Metric" of the target node, i.e. the SGW or PGW, representing the current utilization of the resources as compared to the overall available resources at the target node.

Using the DNS based procedures specified in this document, the node performing the SGW or PGW selection (i.e. the MME, S4-SGSN, ePDG or TWAN), referred to as "selecting node" hereafter, shall prepare the candidate list of the target nodes satisfying the required criteria to serve the new session request.

The criteria for ordering the candidate list of target nodes shall be as specified in clause 4.3.2, i.e. giving priority to collocation when applicable, then to topological matching when applicable, then to the ordering obtained by the S-NAPTR output and then to the ordering obtained by the priority in the SRV records (when SRV records are used). However, the selecting node may select a candidate target node with a lower relative order if all the candidate target nodes with a higher relative order are not available, e.g. when they have reported a Load-Metric=100.

Then within sets of available candidates with the same relative order (e.g. with the same topological order or with the same NAPTR order field) and with the same priority in SRV records (when SRV records are used), further called the "TNodeList" for the list of candidate target nodes "TNode", the node performing the SGW or PGW selection shall use the "Load Metric" and, if available (i.e. when SRV records are used), the DNS-weight-factor of each candidate target node to perform the node selection as indicated below:

1) The selecting node shall calculate the effective available load of each candidate target nodes, considering the current available load and the DNS weight-factor of the target node as follows:

TNode-effective-available-load = (100 – TNode-load-metric)% X TNode-weight-factor

2) The selecting node shall then calculate the relative available load of each candidate target node, representing the effective-available-load of the target node, as compared to the effective-available-load of all the other candidate target nodes as follows:

TNode-relative-available-load = (TNode-effective-available-load / sum-of-effective-available-load-of-all-the-nodes-in-TNodeList) X 100%

3) The selecting node shall then select the SGW or PGW for new session requests according to the TNode-relative-available-load of each target candidate node, e.g. a TNode-relative-available-load value of X% indicates that the corresponding target node should be selected for X% of the new sessions requests.

The selecting node shall assume the following in the calculations above:

– The selecting node shall assume the TNode-load-metric=0 for a candidate target node for which no node level load-metric is available (e.g. due to the load control feature being not supported between two PLMNs and that the PGW and the MME are from different PLMNs);

– The selecting node shall compute the TNode-weight-factor as (65535-NAPTR preference), as per Annex B.2, for each candidate target node if no SRV weight factor is received from the DNS, i.e. when only NAPTR records with the "a" flag are provisioned.

Annex E.1 provides an example illustrating these principles.

When using GTP-C load control, the DNS should not be provisioned with a mix of NAPTR records with an "s" flag and with an "a" flag for a given FQDN.

NOTE: It is not possible to directly compare the TNode-weight-factor used with SRV records with the TNode-weight-factor used when there is no SRV record. It is implementation specific how to distribute the load among candidate SGW or PGWs if the DNS is provisioned with a mix of NAPTR records with an "s" flag and with an "a" flag.

4A.3 APN-level load control

This clause specifies the extensions to the PGW selection procedures which shall be supported when GTP-C load control is supported and enabled at node level and APN level.

A PGW may report the following APN load control information for a given APN via GTP-C signalling (see clause 12.2.4 of 3GPP TS 29.274 [23]):

– APN-Load-Metric: this indicates the current resource utilization for a particular APN, as a percentage, compared to the total resources reserved for that APN at the target PGW; and

– APN-relative-capacity: this indicates the total resources configured for a given APN, compared to the total resources of the target PGW, as a percentage. It is a static parameter and does not change unless the resources configured for the APN change.

The selecting node shall calculate the TNode APN weight factor of each candidate target node, considering the TNode APN relative capacity and the TNode weight factor as follows:

TNode-APN-weight-factor = TNode-weight-factor X TNode-APN-relative-capacity.

The node performing the PGW selection shall then apply the principles and assumption specified in clause 4A.1 with the following additions or modifications:

– the selecting node shall substitute node level (TNode-"X") parameters with their corresponding APN level (TNode-APN-"X") parameters to calculate the relative APN available load of each candidate target node, i.e.:

TNode-APNx-effective-available-load = (100 – TNode-APNx-load-metric)% X TNode-APNx-weight-factor

TNode-APNx-relative-available-load = (TNode-APNx-effective-available-load / sum-of-APNx-effective-available-load-of-all-the-nodes-in-TNodeList) X 100%

– the selecting node shall then select the PGW for new session requests according to the TNode-APN-relative-available-load of each target candidate node, e.g. a TNode-APN-relative-available-load value of X% indicates that the corresponding target node should be selected for X% of the new sessions requests towards the requested APN.

– If the candidate PGW did not provide APN load control information for an APN (called "APNx"), but provided APN load control information for other APNs (called "APNn"), the selecting node shall assume the following when computing the relative APN available load for that candidate PGW for the "APNx":

TNode-APNx-relative-capacity = (100 – sum of [TNode-APN-relative-capacity of other APNs "APNn")]) for the same PGW

TNode-APNx-load-metric = {TNode-load-metric – sum of [for each APNs in APNn (TNode-APN-load-metric X TNode-APN-relative-capacity) / 100]} / {100 – sum of [TNode-APN-relative-capacity of each APNs in APNn]) / 100}

NOTE: The formula to compute the APNx load metric results from the following equation:

TNode-load-metric X TNode-capacity = sum of [for each APNs in APNn (TNode-APN-load-metric X TNode-capacity X TNode-APN-relative-capacity / 100)] + TNode-APNx-load-metric X TNode-capacity X (100 – sum of [ TNode-APN-relative-capacity of each APNs in APNn) / 100

– If the candidate PGW did not provide APN load control information for any APN, the selecting node shall assume the following when computing the relative APN available load for that candidate PGW for an APN:

TNode-APNx-relative-capacity = 100

TNode-APNx-load-metric = TNode-load-metric (i.e. the load at the target node level)

Annex E.2 provides examples illustrating these principles.