← Kubernetes, Slurm & GPU Scheduling / 32

expertNewNVIDIAMicrosoft Azure

Deploy a confidential inference service at an air-gapped site. What must work after the internet is disconnected?

Build a complete import and recovery plan for models, drivers, identity and attestation material. Calculate the disconnected admission window and test fresh-node replacement with external services unavailable.

Updated Nov 2026 · Learn how AI infrastructure works through explanations, worked examples and diagrams, then practise applying it to interview questions.

Build a complete import and recovery plan for models, drivers, identity and attestation material. Calculate the disconnected admission window and test fresh-node replacement with external services unavailable.

The full answer is part of Premium. A free account includes more answers in each topic, but does not unlock this one.

20 answers per topic instead of 10, plus saved progress and bookmarks · no cardor unlock all 324 remaining answers · ₹2,000 / $25

The concepts behind this question

Use these concept explanations to review the ideas behind this question.

🗂️ Scheduling & Orchestration🔒 Premium

Air-Gapped AI DeploymentsAn air-gapped AI deployment needs a controlled import path for software, models and trust material, plus local identity, time, storage and recovery services. Network isolation and confidential execution solve different problems.

🧮 Open Weights & Serving Engines

vLLM Server Arguments That MatterSeparate executor memory, allocated cache pools and logical occupancy. Map context, sequence and token-batch limits to the constraints they control, then tune a pinned deployment against per-rank memory and latency.

🧮 Open Weights & Serving Engines

SGLang Server Arguments That MatterTune static memory, prefill chunks and admission against measured workload behavior. Separate offline throughput heuristics from interactive queue budgets, and calculate the part of request work that prefix caching can save.

🧮 Open Weights & Serving Engines

Model Onboarding: From Hugging Face to ProductionTurn a model release into a reproducible service: pin the complete artifact, confirm the exact engine path, budget transfer and runtime memory, validate request and response behavior, then benchmark and canary against explicit acceptance criteria.

Explore the full curriculum →

UP NEXT ON YOUR JOURNEY

Closest to this topicAn entire rack stops responding at 2 a.m. Walk me through the first thirty minutes.Closest to this topicHow would you measure confidential inference overhead without confusing encryption, cold starts and batching?Closest to this topicDesign an inference service whose operator cannot read customer prompts or the model owner's weights.

FEDITOR'S NOTE

A warm inference process can conceal missing dependencies. The acceptance test must include a clean replacement and expired trust material.

DISCUSSION · 0

No comments yet — be the first to share your approach.