Page 350 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

2 Use Case Description

2�1 Description

In 5G networks, network slicing enables the creation of multiple virtualized networks on shared
physical infrastructure, each optimized for specific service requirements such as latency,
bandwidth, or reliability. However, fluctuating traffic loads across slices can lead to resource
underutilization or congestion, degrading Quality of Service (QoS) for end-users. To address
this, dynamic slice instance management systems leverage machine learning (ML) to monitor
real-time conditions and autonomously adjust resource allocations [1, 2].

The system employs time series analysis to predict traffic patterns, enabling proactive
scaling of slice instances. For example, an ML model might forecast peak usage periods for
an enhanced Mobile Broadband (eMBB) slice streaming high-definition video, triggering
preemptive resource allocation [1]. A load balancer acts as a proxy, distributing traffic across
slice instances and managing their lifecycle—spawning new instances during demand spikes
and decommissioning underutilized ones during low load. This ensures each instance operates
within its capacity thresholds, preventing overloads.

Load balancer allocates a slice instance to the user if the slice instance can accommodate by
analysing the traffic patterns as illustrated in Figure TBD. Load balancer dynamically provisions
new slice instances only if the current instances lack the capacity to accommodate incoming
UE traffic. This decision is triggered by two scenarios:

1. Real-time overload: Existing instances are nearing or exceeding their throughput limits.
2. Predictive scaling: Machine learning forecasts indicate an imminent traffic surge that
would exceed configured capacity thresholds [2].

The load balancer acts as an intelligent proxy, continuously evaluating both live traffic metrics
(e.g., packets/second) and predictive models. When thresholds are breached—whether due to
actual usage or anticipated demand—it automatically spins up preconfigured slice instances.
Conversely, it decommissions underutilized instances during low-traffic periods, ensuring
resource efficiency. This dual-condition scaling ensures uninterrupted QoS while preventing
premature resource allocation and also aligns with global best practices for autonomous
network management, as envisioned in ITU-T frameworks for integrating ML in future networks
[1].
Use Case Status: The implementation plan is in progress.

2�2 Benefits of the use case

This use case enhances the efficiency and resilience of telecommunication infrastructure by
introducing innovative network management strategies. It promotes adaptive and intelligent
network behavior through dynamic slice instance management, supporting the continuous
evolution of telecom systems.

By fostering innovation in network operations, the solution contributes to the creation of more
efficient and resilient infrastructure that can respond to growing demand and operational
complexity.

314