Choose the environment.
Keep the purpose.

Web space, a VM or a container?

These are architectural choices, not a current product-stock list. Confirm the supported environment and its management scope.

Size from evidence.

Describe expected concurrent use, processing behaviour and dependencies before selecting vCPU and memory. Separate persistent storage, temporary files, databases and backup copies. Record network demand and any latency-sensitive connections.

Use representative tests and monitor the result. A resource specification alone does not establish application performance.

Keep the operating boundary clear.

A managed scope should identify precisely what is covered: host, operating system, application, updates, alerts and incident response. Self-management should identify the access and responsibilities retained by your team. Neither label replaces an agreed service description.

Private applications and inference.

For internal tools, APIs or private AI workloads, define identity, permitted data, source access and integration permissions. Confirm model, accelerator and memory requirements against actual capacity before considering deployment.

Prepare the resource brief.

• Application, runtime, operating system and licensing.
• Baseline compute, memory, storage and network observations.
• Expected growth and peak-demand pattern.
• Backup retention, recovery objectives and restore acceptance.
• Required equipment location and access arrangements.
• Named owners for application, infrastructure and incidents.