試験NCA-AIIO トピック1 問題35 スレッド

Real-time AI applications, especially those involving IoT devices, depend on rapid and reliable data ingestion to maintain low latency and high throughput. Intermittent delays and data loss suggest a bottleneck in the network connecting the IoT devices to the processing system. Implementing a dedicated, high-bandwidth network link (e.g., using NVIDIA's InfiniBand or high-speed Ethernet solutions) ensures that data flows seamlessly from distributed IoT devices to the AI cluster, reducing latency and preventing packet loss. This aligns with NVIDIA's focus on high-performance networking for distributed AI, as seen in DGX systems and NVIDIA BlueField DPUs, which offload and accelerate network traffic.
Switching to batch processing (Option B) sacrifices real-time performance, which is critical for this use case, making it unsuitable. A CDN (Option C) is designed for static content delivery, not dynamic IoT data streams, and wouldn't address the core issue of real-time ingestion. Upgrading IoT hardware (Option D) might improve local processing but doesn't solve network-related delays or data loss between devices and the AI system. A robust network infrastructure is the most effective solution here.