workload forecasting

Practical, Bash-first guide to AI VM capacity planning: translate throughput/latency/cost goals into right-sized CPU/RAM/GPU/storage/NIC. Includes rules of thumb for training vs inference (VRAM, KV cache), shell estimators, host surveys, VM shaping (NUMA, hugepages, MIG), repeatable stress/fio/iperf/GPU tests, and KVM/libvirt operations with lightweight monitoring and guardrails to meet SLAs without overspending.