Your organization is operating an edge-to-cloud AI system where hundreds of IoT sensors continuously stream telemetry data to a centralized processing facility. The hosted AI model requires consistent, low-latency, and high-throughput data streams to generate real-time predictive maintenance alerts. Recently, network congestion has caused packet loss and unpredictable transmission delays, leading to degraded model inference accuracy. Which of the following network infrastructure changes would be the most effective solution to restore telemetry stream reliability and performance?
Select an answer to reveal the explanation.
Short Explanation
Here's the deal: if you're streaming real-time IoT telemetry to a central AI engine, congestion is your worst enemy. You need a dedicated, high-bandwidth connection with guaranteed QoS to keep the pipeline clean, fast, and reliable. Caching static content or batching defeats live telemetry.
Full Explanation
In distributed AI environments where telemetry is gathered from edge devices (IoT sensors) and processed centrally, network reliability is paramount. Real-time inference pipelines depend on continuous, predictable data arrival. If the network experiences congestion, it introduces jitter (variable latency) and packet loss. Packet loss means the central AI model receives incomplete data states, directly hurting inference accuracy. Jitter means alerts are delayed, violating real-time SLA requirements. To resolve this, implementing a dedicated, high-bandwidth network link (such as a direct leased line, SD-WAN, or MPLS with strict Quality of Service policies) ensures that the critical AI telemetry traffic is prioritized over general traffic, guaranteeing the required bandwidth and low-latency metrics.
Let's review the distractors: The global CDN option is designed to cache static, read-heavy assets like images or videos near end-users. It does not facilitate real-time, low-latency upload streaming of dynamic telemetry data. The weekly batch upload option would resolve network utilization issues but destroys the core requirement of real-time processing and immediate predictive alerting, converting a real-time system into a delayed historical reporting system. The edge-computing device upgrade option might allow for local processing if the models were deployed at the edge, but it does not address the transmission issues of the current centralized model architecture. Since the prompt states the data must be ingested centrally for the model to deliver insights, the network link remains the critical bottleneck.