SMF supports the cache pod optimization to reduce the cache pod query at the GTPC endpoint. The get affinity query is used to receive the affinity information in an outgoing request or response message toward the GTPC endpoint. With this optimization, the GTPC endpoint pod doesn't send the query to the cache pod for the upcoming request messages.

SMF and cnSGW-C support domain-based user authorization using the Ops Center. To control the access on a per-user basis, use the TACACS protocol in Ops Center AAA. This protocol provides centralized validation of users who attempt to gain access to a router or NAS.

Configure the NETCONF Access Control (NACM) rules in the rule list. Then, map these rules in the Ops center configuration to map the group to appropriate operational authorization. Use the configurations that are based on the following criteria and products:

• With the NACM rules and SMF domain-based group, configure the Ops center to allow only access or update SMF-based configuration.

• With the NACM rules and cSGW-C domain-based group, configure the Ops center to allow only access or update cSGW-C-based configuration.

• With the NACM rules and cSGW-C domain-based group, configure the Ops center to allow only access or update CCG-based configuration.

Note	The NSO service account can access the entire configuration.

To support this feature configuration in Ops Center, the domain-based-services configuration is added in the TACACS security configuration. The TACACS flow change works in the following way:

• If you have configured the domain-based-services parameter, then the configured user name that is sent to the TACACS process, splits user ID into user ID and domain. The split character, which is a domain delimiter, is configured in domain-based-services. These split characters can be "@", "/", or "\" and are used in the following format to get the domain and user ID information.

  • @ — <user id>@<domain>

  • / — <domain>/<user id>

  • \ — <domain>\<user id>

• The TACACS authenticates and authorizes as per the existing flow. However, if the domain-based-services feature is enabled and TACACS authenticates and authorizes the user, following steps are added to the TACACS flow procedure.

  • If Network Services Orchestrator (NSO) logs in as the NSO service account, then that session receives a specific NACM group that you configured in domain-based-services nso-service-account group group-name . This functionally is the same as the way NSO works.

  • If the specified domain exists in the group mapping, then the NACM group that you configured in domain-based-services domain-service domain group group-name is applied.

  • If the user does not have a domain or the domain does not exist in the domain to group mapping, then no-domain NACM group that you configured in domain-based-services no-domain group group-name is applied. If the no-domain configuration does not exist, then the user value is rejected.

To enable this feature, you must configure the domain-based-services CLI command with the following options:

• NSO service account

• Domain service

• Domain delimiter

• No domain

To enable domain-based user authorization using Ops Center, use the following sample configuration:

config 
   tacacs-security domain-based-services [ domain-delimiter delimiter_option | domain-service domain_service_name [ group service_group_name  ] | no-domain group service_group_name | nso-service-account [ group service_group_name | id service_account_id ] ] 
   end 

NOTES:

• domain-based-services [ domain-delimiter delimiter_option | domain-service domain_service_name [ group service_group_name ] | no-domain group service_group_name | nso-service-account [ group service_group_name | id service_account_id ] ] : Configure the required domain-based-services value. The domain-based-services includes the following options:

  • domain-delimiter : Specify the delimiter to use to determine domain. This option is mandatory and allows the following values:

    • @—If domain-delimiter is "@", the user value is in the format: <user>@<domain>.

    • /—If domain-delimiter is "/", the user value is in the format: <domain>/<user>.

    • \—If domain-delimiter is "\", the user value is in the format: <domain>\<user>.

  • domain-service : Specify the list of domains and their group mapping. The key is the name of the domain and group is the group that is assigned to the domain. You must configure at least one option in this list.

  • no-domain : Specify the group that has no domain or if the domain is unavailable in the domain-service mapping, then this group is sent in the accept response.

  • nso-service-account : Specify the NSO service account that has the ID and group. If you configure this parameter, then you must configure the ID and group fields. The ID and group must have string values.

SMF uses the enable-direct-encdec CLI command to optimize the encoding and decoding of the IEs that are associated with the GTPC endpoint pod. This optimization improves the memory management and reduces the garbage collection time.

You can enable this feature at run time. By default this feature is disabled.

To configure this feature, use the following sample configuration:

config 
   instance instance-id instance_id 
      endpoint gtp 
      interface s5e 
         enable-direct-encdec true | false 
      interface s11 
         enable-direct-encdec true | false 
      exit 
   exit 

NOTES:

• enable-direct-encdec true | false : Choose the value as true to enable the encoder and decoder. By default, the value of this field is false .

When large numbers of GTPC peers are connected with SMF or cnSGW-C, the performance of ETCD is impacted. Each peer is a considered as a record in the ETCD, and the timestamp is updated every 30 seconds for each peer. This causes continuous updates on ETCD and generates huge traffic that impacts the overall system performance.

The ETCD Peer Optimization feature facilitates optimization in peer management and enables reduced performance impact on ETCD.

This section describes how this feature works.

Instead of considering each peer as an ETCD record entry, several peers are grouped as a peer group based on the hash value of the IP address of each peer. This reduces the number of entries in ETCD. By default, a maximum of 200 peer groups can be created. For any changes related to a peer in a peer group:

• For a new peer, the peer group is persisted immediately in ETCD.

• For the change in timestamp for existing peers, the peer group is updated once every 3 seconds. This update:

  • Results in a cumulative group update for many peers that have undergone timestamp change within each peer group.

  • Reduces frequent updates to ETCD.

The ETCD client implements the following components:

• Balancer that establishes the gRPC connections to an ETCD cluster.

• API client that sends RPCs to an ETCD server.

• Error handler that decides whether to retry a failed request or switch endpoints.

The App-infra-based pods using memory cache in 5G communicate with the ETCD server. This communication is to store their critical data on a distributed system through the ETCD client load balancer. The memory cache uses the ETCD client 3.3 version, which implements the clientv3-grpc1.7 library.

SMF supports the ETCD client, version 3.4.21, which implements the clientv3-grpc1.14 library. This library is backward compatible.

The clientv3-grpc1.14 library has the following key points:

• Simplify the balancer failover logic instead of maintaining a list of unhealthy endpoints.

• Create multiple subconnections, which implies one subconnection per endpoint, when multiple endpoints are available. For example, in a five-node cluster, clientv3-grpc1.14 balancer requires five TCP connections.

• Provide more flexible load balancer with a better failover performance by preserving the pool of TCP connections.

• Extend the default round robin balancing policy to support other types of balancers, such as power of two and pick leader.

• Use the gRPC resolver group and implement balancer picker policy to delegate complex balancing work to upstream gRPC.

• Implement retry in the gRPC interceptor chain that handles gRPC internal errors automatically and enables advanced retry policies, such as backoff.

After receiving an error for a PUT request, the ETCD client retries this request thrice. The ETCD client performs each retry after a timeout of one second. The ETCD client completes all the three retries within three seconds, which meet the existing SLA with the applications. After three retries, the ETCD client returns the outcome to the application.

SMF updates the CDL when the subscriber state changes from idle to active, and when the ULI, UeTz, or the serving network is modified.

When the transaction requests driven to CDL increases, SMF incurs a higher CPU utilization. To prevent the needless CPU utilization, SMF updates only a subset of the CDL with the changed attributes.

Flag DB database is updated for the following SMF procedures:

• MBR with only ULI change—SMF handles MBR with only ULI change, in a stateless way to send the response. After sending the response, the smf-service updates the CDL, which impacts the CPU utilization. To optimize the CPU usage, SMF notifies the CDL about the ULI only with the partial updates.

• 4G RAT Handover—During an inter S-GW handover, the smf-service receives the MBR with a ULI change and the TEID change. To optimize the CPU usage, SMF notifies the CDL about peer TEID and ULI only with the partial updates, if the handover is successful for all the bearers.

• N2 Handover—When the N2 handover procedure ends, the smf-service updates the CDL which impacts the CPU utilization. To optimize the CPU usage, the SMF notifies the CDL about only the ULI and TEID with the partial updates, if the handover is successful for all the existing QFI.

The Radio Resource Control (RRC) is a layer within the 5G NR protocol stack. It exists only in the control plane, in the UE, and in the gNB. The existing state of PDU sessions controls the behaviour and functions of the RRC.

During the PCF-initiated modification, the AMF sends those received unsuccessful transfer radio networks cause codes in the N2 content of the SmContextUpdate message to the SMF under the following conditions:

• When the Xn-handover is in progress.

• When the AN is released.

• When the UE is in the RRC inactive state.

• When the UE isn’t reachable.

The SmContextUpdate message with the N2 cause codes acts as a bridge and converter from AMF to SMF or from SMF to AMF.

This section describes how this feature works.

During the PCF-initiated modification, when the Xn-handover is in progress, the following scenarios are noted:

• The AN gets released or the UE is in RRC inactive state and not reachable.

• The AMF relays the received unsuccessful transfer radio network cause code, in the N2 content of the SmContextUpdate message to SMF.

• These cause codes could be standard radio network causes or there could be some customized radio network cause codes being sent from the gNB.

Previously, these cause codes were rejected by the SMF and the PCF was attempting multiple times the same PCF modifications. Now, the SMF doesn't reject immediately, and behaves differently for different cause codes, based on the new N2 trigger configuration to avoid multiple reattempts from the PCF.

The following scenarios are supported in the SMF for the PDU Modify procedure, based on the received N2 cause code:

• When the cause code indicates that the Xn-handover is in progress or the AN gets released, then the following activities occur:

  • The SMF suspends the ongoing PDU session modification.

  • It resumes back after the Xn-handover or the AN Release.

• When the cause code indicates that the UE is RRC inactive and not reachable, then the following activities occur:

  • The SMF rejects the PDU session modification.

  • It reports the rule failure to the PCF.

By default, this feature gets activated for a few standard RRC inactive cause codes with default guard timeout and zero max-retry.

For the following cause codes, the SMF suspends session modification, and resumes only after the Xn-handover activity gets over:

• _RadioNetwork_NG_intra_system_handover_triggered

• _RadioNetwork_NG_inter_system_handover_triggered

For the following cause codes, the SMF rejects the session modification:

• _RadioNetwork_UE_in_RRC_INACTIVE_state_not_reachable

Along with the standard cause codes, a new N2 trigger CLI is introduced to configure the different customized radio network cause codes, and the corresponding SMF actions.

Note	The non-roaming PCF-initiated modification scenarios are supported as a part of this feature.

To configure this feature, use the following sample configuration:

config 
profile access access_profile_name 
    n2 trigger { ho-in-progress | temp-not-reachable } { guard-timeout timeout | max-retry retry_count | value retry_count_range } 
    n2 trigger ue-not-reachable value notreachable_count_range 
    end 

NOTES:

• profile access access_profile_name —Specify a name for the access profile.

• n2 trigger —Specify the N2 trigger type. Trigger can be the traffic type. Must be one of the following:

  • ho-in-progress —Specify the handover-in-progress trigger configuration list of cause-codes.

  • temp-not-reachable —Specify the temporary not reachable trigger configuration list of cause-codes.

  • ue-not-reachable —Specify the UE not reachable trigger configuration list of cause-codes.

• guard-timeout timeout —Specify the Handover in progress guard timeout in milliseconds, within the range of 500-30000 milliseconds. The default value is 10000 milliseconds.

• max-retry retry_count —Specify the maximum retry count value for the handover in progress or temporary not reachable options, within the range of 0-64. The default value is 0.

• value { notreachable_count_range } | { retry_count_range } —The numbered value in the range of counts for UE not reachable or the maximum retry range.

Note

The defined configurations are used to match the received unsuccessful transfer (radio network standard and customized) causing the code to decide the RRC inactive action. It has the following scenarios: The PCF-initiated modification gets rejected in the case of ue-not-reachable , ho-in-progress , and temp-not-reachable cases. It gets suspended and resumes back after the xn-handover activities in the AN release. The Guard Timer gets started, when the RRC inactive action is either in the ho-in-progress or the temp-not-reachable trigger profile. It waits for the ongoing PCF-initiated modification to suspend. It restarts the xn-handover activities in the AN release within the given time. If this action fails, then the PCF-initiated modification gets rejected. As a result, this action reported as the rule failure note to the PCF. The maximum retry allows the maximum continuous reattempt after the first attempt gets failed. It's a result of receiving the same trigger category cause code repeatedly for each and every attempt. If this action fails, the PCF-initiated modification gets rejected. As a result, this action reported as the rule failure note to the PCF, after reaching the maximum retry attempt. This feature gets activated for the standard RRC inactive cause codes with a default guard timeout and zero maximum-retry.

The Resiliency Handling feature introduces a CLI-controlled framework to support the service pod recovery, when you observe a system fault or a reported crash. It helps in recovering one of the following service pods:

• sgw-service pod

• smf-service pod

• gtpc-ep pod

• protocol pod

• diameter-ep pod

These service pods are software modules containing the logic to handle several session messages. The service pods are fault-prone due to any one of the following or a combination of multiple scenarios:

• Complex call flow and collision handling

• Inconsistent session state

• Incorrect processing of inbound messages against the session state

• Unexpected and unhandled content in the inbound messages

Whenever you observe the system fault or a crash, the fault behavior results into a forced restart of the service pod. It impacts the ongoing transaction processing of other sessions. The crash reoccurs even after the pod restart.

To mitigate this risk, use the CLI-based framework with actions defined to clean up subscriber sessions or terminate the current processing.

This section describes how you can use the fault recovery framework to define actions for the crash. The framework allows you to define any of the following actions:

• Terminate—When a fault occurs, this action terminates the faulty transactions, and clears the subscriber session cache. It’s applicable for smf-service and sgw-service pods.

  Note	The pod doesn't get restarted. The database doesn't get cleared during this action.

• Cleanup—When a fault occurs, this action clears the faulty subscriber session and releases the call. It’s applicable for smf-service and sgw-service pods.

• Graceful reload—When a fault occurs, this action restarts the pod. It’s applicable for gtpc-ep, protocol, and diameter-ep pods. It handles the fault signals to clean up resources, such as the keepalive port and closes it early. It also allows the checkport script to detect the pod state and initiates the VIP switch processing for the corresponding pods.

• Reload—When the pod crashes, it initiates the reloading activity. It’s a default setting or value applicable for all the pods.

To configure this feature and to enable the system fault recovery, use the following sample configuration:

config 
    system-diagnostics { diameter |  gtp | pfcp | service | sgw-service } 
        fault 
            action { abort | cleanup { file-detail | interval | num | skip { ims | emergency | wps } } | graceful-Reload | reload } 
            end 

NOTES:

• system-diagnostics { diameter | gtp | pfcp | service | sgw-service } —Specify the required type of service pods for system diagnostics. The available pod options are diameter, gtp, pfcp, smf-service, and sgw-service.

• fault —Enables fault recovery while processing sessions.

• action { abort | cleanup | graceful-Reload | reload } —Specify one of the following actions to take on fault occurrence. The default action is reload.

  • abort —Deletes the faulty transaction and clears its session cache. The database doesn't get cleared.

    Note	It's an exclusive option to the smf-service pod.

  • cleanup { file-detail | interval | num | skip } —Enable the cleanup activity. It has the following selections to mitigate the fault action:

    • file-detail —Lists the file names with line numbers. It excludes the file name details from the recovery.

    • interval —Specifies the duration of the interval in minutes. This duration specifies the permissible interval within which it allows the maximum number of faults. Must be an integer in the range 1–3600.

    • num —Specifies the maximum number of tolerable faults in an interval. Must be an integer in the range 0–50.

    • skip { ims | emergency | wps } —Enable the skip cleanup of a subscriber session for an active voice call, or the WPS, or an emergency call.

      • To detect the active voice calls, use the following command:

        profile dnn dnn_name ims mark qci qos_class_id 

      • When you enable the skip cleanup configuration, the SMF deletes the faulty transaction, and clears its session cache.

      • When a fault occurs during the session setup or the release state, the SMF performs the following:

        – Deletes the transactions on the session end.

        – Overrides the configured fault action during these states.

        – Clears the session cache and database entries for the faulty transaction.

      • It allows the dynamic configuration change.

    Note	It's an exclusive option to smf-service and sgw-service pods.

  • graceful-Reload —Specify the option to gracefully reload the pod. The service pod handles fault signals to clean up resources like the keepalive port and continues with crash processing (pod restart processing).

    Note	It's an exclusive option to diameter-ep, gtpc-ep, and protocol service pods.

  • reload —Reloads the pod, when it crashes due to a faulty behavior. It's an option applicable to all the service pods. It’s also the default option.

This section describes operations, administration, and maintenance support for this feature.