5G Core Deployment Optimization with Rebaca’s Testing Solution

Rebaca Technologies is a 2024 Nominee for the TECHCOnnect Innovation Showcase. Learn More Here.

The 5G ecosystem is not just limited to Network Functions (NFs). It also includes the underlying virtualized infrastructure (NFVI) layers. CSPs need to optimize their multi-vendor 5G core network deployments on infrastructure using VMware and Intel technology. CSPs have to validate and right-size a use case-specific architecture to reduce costs and increase efficiency. The ability to create real-life scenarios in the lab and capturing KPIs from multiple layers (NFVI, NFV, mobility and application) significantly helps in a right-sizing exercise.

The goal is to utilize the underlying infrastructure from Intel and VMware and support use cases resiliently at optimal costs and efficiency, which can contribute to a sustainability initiative.

There are a few basic requirements to meet this goal:

• Generate the deployment-specific call models for both Control and Data plane scenarios from real-life network data.
• Validate successful interoperability of the deployment architecture in an automated fashion through a CI/CD workflow.
• Collect and analyze KPIs from different layers and correlate them for optimal sizing and efficiency.
• Work with the SMO from VMware for dynamic sizing automation

ABot, the flagship product of Rebaca Technologies, is a cloud-native 4G/5G Continuous Test Assurance and Analysis solution. It has 3GPP-compliant test cases in simple Domain Specific language (DSL), and the solution that can emulate any 4G/5G or open RAN NF, generate real-life control plane and data plane traffic, and analyze execution results to validate NFs acting as a device under test (DUT). It collects and displays real-time KPIs for both infrastructure components and 3GPP-based NFs by processing packet captures, log messages, and infrastructure data. It can empirically validate workloads for scale and performance and demonstrate elasticity in an environment with technology from VMware and Intel.. This unique capability helps compute optimal 5G core deployment footprints for cost optimization and verify scalability of the solution with the underlying infrastructure from VMware and Intel.

Real-life Usage of ABot in a Multi-Vendor 5G Core Deployment

The usage of ABot in optimizing multi-vendor 5G core deployments is not just a concept. It is already being used actively in labs and field trials to validate and optimize 5G core deployments on cloud platforms. For example, a leading Korean operator, instead of using a monolithic high-end 5G core, is evaluating the combination of control plane NFs of a vendor along with ultra-high-throughput UPF (in TBPS) of a specialized ISV. ABot is the chosen solution to validate interoperability of this combination for different functional and load scenarios. In another case, a Fortune 500 server vendor is using ABot in their interoperability telco lab for validating multi-vendor NFs on various infra-block under different deployment configurations.

Demo Setup for the Proof of Concept

The ABot Team has prepared a demo setup consisting of ABot deployed along with a 5G core on a hybrid environment with VMware and Intel technology. The ABot deployment comprises its Test Execution engine, the Analytics module and the AI/ML module. The Test module includes the execution framework, as well as the 3GPP stack nodes (emulators). The Analytics module facilitates the processing of KPIs from different sources and their consolidated visualization. The AI/ML module (ABot-AI) uses deep learning to generate predictions of system behavior and offer recommendations.

In this setup, ABot emulates the UE and gNodeB(s) and generates different levels of control plane and data plane load on the 5G core components. The 5G core deployment is based on specified resource requirements for a defined level of load. The POC is initiated by executing a set of functional tests through ABot against the deployed 5G core. The visual test results offer a clear insight on the functionality of the deployed 5G core. Any test failures are associated with detailed Root Cause Analysis (RCA) indications and recommendations that assist in troubleshooting and fixing the underlying issues.

The next step in the POC is to execute load tests through ABot against the deployed 5G core. The resulting infrastructure, mobility and application KPIs show the response of the system for the generated load. The load is now increased gradually to the defined maximum level. If the mobility KPIs (such as control plane latency) remain well within acceptable limits, it may be implied that the system resources are overestimated and could be reduced. In this situation, the system resource allocations are reduced, and the test is repeated. Alternatively, if KPI degradation is observed before reaching the specified load limits, the system resource specifications will need to be revised upwards. Ultimately, you can identify the minimum system resource allocation at which the mobility KPIs remain at acceptable limits for the maximum load that the 5G core is designated to handle. The result:

The adjusted system resource allocation is optimally sized for the given load level. 

The multi-step test execution and modification in resource allocation of the deployed system can be orchestrated through automation.

Potential Usage of the AI/ML module to Predict System  Behavior Under Load

Replicating production call models and validating the target solution architecture against it is essential. Production-level network data needs to be used to determine the call models in the deployment scenario. ABot-AI is being optimized to automatically identify end-to-end 3GPP procedures from a network trace. It can further identify anomalies or predict which configurations will be optimal.

Load characteristics in terms of traffic arrival are dynamic in nature. Resources like memory and processors are in heavy demand during load and cause latency. The throughput of the resources and the versatile demand of the traffic are variable. Thus, a linear predictor for resource sizing in a load scenario is not optimized.

ABot-AI is multi-variant in nature and uses AI-ML techniques to learn the dynamic behavior of the system for different traffic and load characteristics and their trends for various resource configurations. Based on its continuous learning, ABot-AI predicts the right resource requirement for keeping the latency and efficiency at the recommended levels.

---