A social media company wants to expand its agentic system to support global users, minimize...

A global user base requires regional placement, failover, and dynamic allocation. Docker alone packages the app; it does not solve cross-region availability. The correct implementation surface is separate scalable inference services with load balancing, readiness checks, and resource policies tied to latency and GPU metrics. The selected option specifically B states “Designing a distributed system architecture with multi-region deployment, automated failover, and dynamic resource allocation”, which matches the operational requirement rather than a superficial wording match. In a GPU-backed agent deployment, Option B maps closest to how the NVIDIA stack expects orchestration, inference, and control policies to be separated. The alternatives would look simpler in a prototype, but placing all roles on the same large model wastes GPU budget and makes simple requests pay the cost of complex reasoning. This lines up with NVIDIA guidance because Triton can expose request and GPU metrics while Kubernetes policies translate those signals into scheduling and autoscaling decisions. This choice gives engineering teams the knobs they need for continuous tuning after deployment.