Week 11 · Lesson 5

On-Premises vs Cloud

When you finish specifying a server, the next question is where it runs — in your own building, or in someone else’s? The answer is not a preference; it is a decision with cost, speed, latency, and security consequences that you have to be able to name precisely.

CCNA 200-301 CyberOps 200-201 NIST SP 800-145 Verified 2026-10-11
Part 1

What cloud computing is

The canonical definition comes from NIST SP 800-145: cloud computing is a model for enabling convenient, on-demand network access to a shared pool of configurable computing resources that can be rapidly provisioned and released with minimal management effort. The key words are on-demand, shared pool, and rapid provisioning — a cloud is not just “someone else’s server.”

Definition — on-premises. On-premises (on-prem) means the server hardware lives in the organisation’s own facility and the organisation owns and operates it end to end — the traditional model the rest of Week 11 has described.

NIST lists five essential characteristics, and exams like crisp phrasing:

The five essential characteristics · select one

Exam tip. If a scenario describes a provider’s data centre, a shared pool of virtual machines, and self-service provisioning, it is describing the five characteristics — not merely “a remote server.” “Someone else’s server” is only one part of the definition.
Part 2

Service models: IaaS, PaaS, SaaS

The three service models describe how much of the stack the provider runs. The higher you go, the less you manage — and the less control you hold.

ModelProvider gives youYou manageExample
IaaSVirtual machines, storage, and networking — raw infrastructureThe OS and everything above itAWS EC2 virtual servers
PaaSA managed runtime and platform to deploy ontoYour application and its dataA managed app/database platform
SaaSA complete, ready-to-use applicationOnly your data and settingsWeb-based email or CRM
Service model explorer · select a model

Highlighted rows are run by the provider; plain rows are yours to run and patch.

Exam tip. Anchor the stack by where management starts: IaaS you manage from the OS up; PaaS you manage from the application up; SaaS you manage nothing. If a scenario says “we install and patch the operating system,” that is IaaS. If it says “we just write and deploy our code,” that is PaaS. If it says “we just log in and use it,” that is SaaS.
The classic example. AWS EC2 is the textbook IaaS: you rent a virtual machine and own the OS image, patches, and sizing. Stacked on top, a managed database or application runtime is PaaS, and a browser-delivered email or office suite is SaaS.
Part 3

Deployment models: public, private, hybrid

Deployment models describe who the cloud serves:

Deployment model picker · select a model
On-premises Your own hardware You own and operate everything end to end. Cost: CAPEX up front, then fixed running costs. Not a cloud deployment model Private cloud Dedicated to one organisation Exclusive use; most control and isolation. On your premises, or hosted for you alone. Public cloud VM · Storage · Network Open use by the general public; shared pool. Cost: OPEX, billed like a utility. AWS · Azure · GCP hybrid

Remember the wording. NIST defines public as “provisioned for open use by the general public” and private as “provisioned for the exclusive use of a single organisation.” Those two phrases are the exam answer; a hybrid is simply two or more distinct infrastructures bound together.
Part 4

On-premises versus cloud: the dimensions

Compare the options on the axes that decide real projects:

DimensionOn-premisesCloud
Cost typeCAPEX: buy hardware up front, then depreciate itOPEX: rent monthly; no capital purchase
Provisioning speedWeeks to order, rack, and configureMinutes, self-service
ScalingBuy bigger hardware (vertical); slowAdd/remove instances (horizontal); near-instant
LatencyVery low: on-site, one local hopAdds a round trip to the provider’s region
Availability / DRYou build resilience; a second site is costlyProvider regions and zones offer resilience by design
Security responsibilityEverything is yoursShared with the provider (Part 5)
Cost predictabilityFixed once purchased; deprecation on a scheduleVariable month to month; can surprise if unmanaged
Definition — vertical vs horizontal scaling. Vertical scaling (scale up) means making one machine bigger: more CPU, more RAM. Horizontal scaling (scale out) means adding more machines. Cloud favours horizontal scaling because instances are cheap to create and destroy.
Vertical scaling — make one machine bigger One server More CPU More RAM bigger Horizontal scaling — add more machines Load balancer VM 1 VM 2 VM 3 Add or remove instances in minutes
Vertical scaling is a one-machine upgrade with a hard ceiling and a real hardware lead time. Horizontal scaling adds cheap, disposable instances — the reason the cloud can absorb a traffic spike that would take months of procurement to fix on-premises.
Latency is the quiet decider. Bandwidth is easy to buy; physics is not. Any workload that needs a sub-millisecond, guaranteed round trip — factory control, high-frequency trading, real-time voice processing — is usually kept on-premises or at the edge. Cloud round-trip latency is fine for web and business applications but is the wrong fit for hard real-time control loops.
Part 5

The shared responsibility model

In the cloud, security is shared: the provider secures the layers it owns, and the customer secures the layers above. This is the shared responsibility model, and the split moves with the service model:

ModelProvider securesYou secure
IaaSPhysical, host, hypervisor, network fabricOS, runtime, application, data, identity
PaaSEverything through the runtime and platformApplication, data, identity
SaaSThe entire application stackYour data, access settings, and users
The exam trap. “Moving to the cloud” does not move all security to the provider. If you run an EC2 instance (IaaS) and never patch its operating system, that misconfiguration is your responsibility, not the provider’s. The provider secures of the cloud; you secure what you put in the cloud.

Read the table one more time from the bottom up. As you move from IaaS to PaaS to SaaS the provider’s column grows and yours shrinks — but it never reaches zero. Even with SaaS, the provider secures the application while you still own your data, your accounts, and who is allowed to log in. The service model decides how much you must manage, not whether you must manage it.

Part 6

Virtualization: the enabler

The cloud could not exist without virtualization — the abstraction of physical hardware into software-defined resources. One physical host runs a hypervisor that creates many isolated virtual machines, each believing it owns a complete computer.

Physical host A Hypervisor — resource pooling and isolation VM tenant 1 VM tenant 2 VM tenant 3 VM tenant 4 One host serves many customers safely — that is multi-tenancy. Physical host B Hypervisor — ready to absorb the moved VM VM tenant 5 VM tenant 6 A running VM can move here with no downtime. Live migration — rebalance load, evacuate a failing host, patch without interruption
Virtualization is the enabler underneath every cloud characteristic. Multi-tenancy makes the economics work; live migration makes maintenance invisible; the same mechanism creates and destroys instances in seconds, which is rapid elasticity.
Connects to the rest of the course. Virtualization is the same idea as the server roles in earlier lessons: one physical box serving many logical roles. At cloud scale, software (not extra hardware) generates those roles.
Part 7

Decision matrix

Use this to place a workload. Read the left column, follow the preference:

If the workload…Lean towards
Needs sub-millisecond, deterministic latencyOn-premises / edge
Has unpredictable or spiky demandCloud (elastic, horizontal scaling)
Is a standard app with modest, steady loadEither; compare TCO
Must ship fast with a small teamCloud (PaaS/SaaS), no hardware lead time
Handles highly sensitive, regulated data with strict localityPrivate cloud / on-prem
Needs best-effort DR without a second siteCloud (multi-region)
Is a legacy app tied to specific hardwareOn-premises
Workload placement adviser · choose the workload

Key terms

Part 8 · Your turn

Review

Check what you have learned with these review questions.

Review · 6 questions

Sources

Sources & references

  1. Mell, P. & Grance, T. NIST SP 800-145: The NIST Definition of Cloud Computing (2011) — authoritative source for the five characteristics, three service models, and four deployment models. csrc.nist.gov.
  2. Microsoft Azure. Shared responsibility in the cloud. learn.microsoft.com.
  3. Amazon Web Services. What is Cloud Computing? and Types of Cloud Computing (IaaS, PaaS, SaaS). aws.amazon.com.
  4. Microsoft Learn. Azure IaaS, PaaS, SaaS definitions. learn.microsoft.com.
  5. Cisco. CCNA (200-301) Exam Topics, virtualisation and cloud fundamentals. cisco.com.

Target exams: CCNA 200-301 v1.1 · CyberOps Associate 200-201. Last content verification: 2026-10-11.