How-to/ 5 min read Inference is not training: sizing for serving, not for the run Training is throughput-bound and batch-friendly. Serving is latency-bound and bursty. Sizing one like the other is how GPU bills get strange. Sep 2, 2026 How-to/ 6 min read Sizing a GPU instance: what to put around the accelerator A GPU is only as fast as the CPU, memory and disk feeding it. How to size the rest of the instance so the accelerator is the bottleneck. Sep 2, 2026 How-to/ 6 min read Ingress, LoadBalancer or NodePort: exposing a service without guessing Three ways to get traffic into a Kubernetes cluster, what each one actually costs, and the decision rule that stops you paying for twelve load balancers. Aug 31, 2026 How-to/ 5 min read Upgrading a production Kubernetes cluster without downtime The pre-flight checks, the PodDisruptionBudgets that make a node drain safe, the upgrade order, and why there is no rollback for a control plane. Aug 31, 2026 How-to/ 5 min read Persistent volumes on managed Kubernetes: what to use for what Block, file and object storage in a cluster: which primitive fits which workload, the access-mode rules that surprise people, and the reclaim policy that deletes production data. Aug 31, 2026 How-to/ 5 min read A working 3-2-1 backup setup with Veeam A concrete Veeam configuration implementing 3-2-1-1-0 against object storage with immutability, plus the retention arithmetic and the settings people get wrong. Aug 23, 2026 security/ 4 min read The 30-minute hardening baseline for a public VM The short list of changes that removes almost all opportunistic attacks against an internet-facing Linux VM, with the commands to apply them. Aug 23, 2026