IaaS is like renting an empty warehouse: the provider supplies physical hardware, power, cooling, and networking, while you furnish and maintain the virtual machine. You choose its operating system, install runtimes, configure network drivers, and deploy your code. section 9.4.1
That control comes with a hands-on maintenance bill:
- You get root access for custom operating-system settings, background services, or legacy software. section 9.4.1
- You handle machine patching and maintenance; virtual machines are billed by the hour, even when no users are active. section 9.4.1
- Consider IaaS when a project needs custom OS-level drivers, extreme control, or a fast migration of a legacy server. For a new project, weigh that flexibility against the ongoing operational work. section 9.4.1 section 9.4.6
Table 9.1 Architect’s Cheat Sheet: The Compute Spectrum
|
Model
|
What It Is
|
Control Level
|
"Clarity" Best Use Case
|
The Hidden Architectural Trap
|
|
IaaS (Virtual Machines)
|
Renting the metal.
|
High (OS-level access)
|
Legacy applications or strict legal/compliance
requirements.
|
"Lift and Shift":
Moving legacy problems to the cloud and paying 24/7 for idle time.
|
|
CaaS (Containers)
|
Renting the shipping port.
|
Medium (Environment &
runtime)
|
Standardized, portable deployments across different
environments.
|
Orchestration Overhead: You
still have to manage the complex system (Kubernetes) and route the boxes.
|
|
PaaS (Platform as a Service)
|
Renting the runtime.
|
Medium-Low (App code &
config only)
|
Rapid prototyping and standard web apps, where you want
zero infrastructure chores without learning Docker.
|
The "Black Box": If
your app suddenly needs a custom system-level library, you are out of luck
and heavily locked into that vendor's ecosystem.
|
|
FaaS (Serverless)
|
Renting the millisecond.
|
Low (Pure logic/code)
|
Spiky traffic or new startups wanting infinite scale with
zero idle costs.
|
"Cold Starts" and Time Limits: Latency
spikes when waking up code, and hard execution limits that brutally kill
long-running processes mid-flight.
|
|
AIaaS (AI Models)
|
Renting the reasoning.
|
Zero (Prompts & payload)
|
Adding complex language or data processing without
building ML infrastructure.
|
Hallucinations: The model
confidently invents facts, requiring strict "RAG" guardrails.
|
Compare IaaS with other compute models by control level, best-fit use case, and architectural trap. section 9.4.6







