Workstation 博客
Observability
标记为 Observability 的文章。
NVIDIA Minerva for AI Factories: Turning Alert Storms into Recovered GPU Hours
One degrading optic can raise thousands of alerts and stall a thousand GPUs. What NVIDIA Minerva does, what it is worth in GPU hours, goodput, MTBI, MTTR and tokens, which of its three consumption models fits, worked examples for a Vera Rubin NVL72 neocloud, a university Blackwell cluster and an enterprise with Grafana and ServiceNow, and how it relates to UFM.
Balinder Walia阅读 
NVIDIA UFM for AI Clusters: Why InfiniBand Fabric Visibility Decides What Your GPUs Are Worth
One flapping InfiniBand link can stall a 256-GPU job. What NVIDIA Unified Fabric Manager does, when UFM Telemetry, Enterprise or Cyber-AI fits, worked examples for a 4-node DGX B300 cluster, a 32-node SuperPOD-style cluster and a university Slurm service, and what fabric visibility is worth in GPU hours.
Balinder Walia阅读 
Observability for Agentic AI: Tracing, Evaluating and Auditing Agents
AI agents fail quietly: a confident answer from the wrong document, a tool called with the wrong argument. Learn how to trace every model, retrieval and tool step with OpenTelemetry, attach quality scores, and keep the records auditors ask for.
Beyond Unit Tests: Testing How Your System Fails and Recovers
Unit tests prove a function is correct, not that a system survives a slow database or a lost zone. A practical guide to cold-start integration tests, fault injection with Toxiproxy, circuit breakers, recovery drills and safely automated mitigation.
Agentic AI in Production: Permissions, Approval Gates, Traces and Evals
A practical playbook for running AI agents in production: classify every tool by side effect, gate mutations behind a human, set hard stop conditions, trace each step with OpenTelemetry and evaluate continuously.
Observability Is Not Optional: Make It a Business Standard
Why fragmented monitoring creates blind spots, duplicate spend, and slow incidents — and how to make centralised observability a non-negotiable company standard.
Balinder Walia阅读 
揭示 LLM 瓶颈：可观测性、OTEL 和成本控制
Workstation 通过可观察性揭示 LLM 代理瓶颈：OpenTelemetry、Prometheus/Grafana/Thanos、Langfuse — 具有明确的优缺点、成本杠杆和保护利润的使用指标。
Balinder Walia阅读 
