Deployment models and cloud architecture
Clear study notes on cloud deployment models (public, private, hybrid, community, multi-cloud) and cloud architecture (components, reference architecture, availability formulas, CAP theorem), with definitions, derivations, diagrams to draw, worked examples, exam guidance and quick checks.
Drafted with Aria, reviewed by the AiCanCode.org team · Updated . How we write and check our notes. Spotted an error? Use Give Feedback at the bottom of the page.
Why it matters
Cloud deployment models and cloud architecture tell engineers where to place applications and data, and how to organize components (clients, load balancers, virtual machines, storage, networks) to meet requirements for cost, security, performance and reliability. Choosing the correct deployment model (for example Public cloud vs Private cloud vs Hybrid cloud) and designing the cloud architecture are common tasks when migrating systems to the cloud, building cloud-native apps, or planning disaster recovery.
Key ideas
Start by defining the basic terms (every bolded term is defined when first used):
Also know the cloud service models commonly examined:
- IaaS (Infrastructure as a Service): provider offers virtualized compute, storage, and networking; customer manages OS, runtime, data, and applications.
- PaaS (Platform as a Service): provider manages OS, runtime, and middleware; customer deploys applications and manages data.
- SaaS (Software as a Service): provider delivers the full application; customer manages only data inputs and configuration.
- FaaS/Serverless: provider executes functions on demand; customer supplies code and configuration; billing per invocation/runtime.
And the shared responsibility model: the provider secures the cloud (facilities, hardware, and managed service internals), while the customer secures what they put in the cloud (identities, OS/configuration for IaaS, data classification, access controls, encryption, and network policies).
Cloud deployment model: a specification of how cloud infrastructure is provisioned and who has access. It determines ownership, management, and network access.
Cloud architecture: the set of components (clients, services, middleware, storage, network, virtualization layer) and their interactions that implement cloud services.
Public cloud: a deployment model where the cloud infrastructure is owned and operated by a third-party provider and resources are shared among multiple tenants over the public Internet.
Private cloud: a deployment model where the cloud infrastructure is operated solely for a single organization; it can be on-premises or hosted by a provider but provides dedicated resources.
Hybrid cloud: a deployment model that composes two or more distinct clouds (private, public, or community) bound together by standard or proprietary technology that enables data and application portability (for example, cloud-bursting for load spikes).
Community cloud: a deployment model shared by several organizations with common concerns (security, compliance) — e.g., several hospitals sharing a healthcare cloud.
Multi-cloud: an operational approach using multiple cloud providers in parallel for redundancy, best-of-breed services, or vendor lock-in avoidance.
Virtualization layer (hypervisor): the software layer that allows multiple virtual machines to share physical hardware; containers use OS-level virtualization (container runtime) sharing the host OS kernel.
Front-end and Back-end (in cloud architecture): the front-end is the user-facing components (clients, web browsers, mobile apps); the back-end comprises servers, storage, and control systems that deliver services.
Load balancer: a network component that distributes client requests across multiple servers to improve throughput and availability.
SLA (Service Level Agreement): a contract specifying guaranteed performance and availability targets between provider and consumer.
Availability: the fraction of time a system is operational and accessible. Defined precisely below with formulas.
Exact statements and theorems commonly used in exams:
CAP theorem (Brewer's theorem, exact exam-style statement): In a distributed data system that can suffer network partitions, it is impossible for the system to simultaneously provide all three guarantees: Consistency, Availability, and Partition tolerance. Assumption: the system must continue operating in presence of arbitrary network partitions; choice is which of the other properties to sacrifice.
- Consistency: every read receives the latest written value (linearizability/strong consistency).
- Availability: every request receives a (non-error) response, without guarantee it is the latest value.
- Partition tolerance: the system continues to operate despite arbitrary message loss or delay between nodes.
Formal definition of Availability used in operations engineering (exact): Availability = uptime / (uptime + downtime). Under steady-state modelling with repairable components,
- Define MTBF (Mean Time Between Failures): expected time the system runs before a failure (units: time).
- Define MTTR (Mean Time To Repair): expected time to restore service after a failure (units: time).
From these definitions, derive Availability (A):
Step 1: One failure cycle length = MTBF + MTTR. Step 2: Fraction of time operational = MTBF / (MTBF + MTTR). Therefore Availability A = MTBF / (MTBF + MTTR). The assumption is exponential or memoryless failures are not required for the formula; it is a time-average definition.
Architecture diagrams to draw (exam instruction style):
Hybrid cloud for an e-commerce site: Draw from left to right.
- Draw small rectangles labelled "Client (Web/Mobile)" connected by an arrow to a box labelled "Internet" then arrow to a box "Load Balancer (Public)".
- From Load Balancer draw arrows to two boxes inside a dashed cloud labelled "Public Cloud: Web & App Tier (auto-scaled VMs)". Label these VMs and show multiple instances.
- Also draw a secure VPN line down to a separate dashed box labelled "Private Cloud / On-prem DB" containing "Database (Private subnet)" and "Backup storage". Place a firewall icon on the boundary and label "VPN / Direct Connect".
- Add arrows from Web/App Tier to a managed object storage box in the public cloud labelled "Object Storage (S3-like)" and to a CDN at the edge.
- Label components: Load Balancer, Auto-scaling group, Public subnets, Private subnets, NAT gateway, Security groups, Monitoring/Logging.
N-tier cloud reference architecture: Draw tiers: Presentation -> Application -> Data -> Storage/Backup, with load balancers between presentation and application tiers, and DB replicas across availability zones. Mark replication arrows and failover paths.
Formulas
A = uptime / (uptime + downtime)
- A: availability (dimensionless, often expressed as percentage). Applies to measured observation windows.
A = MTBF / (MTBF + MTTR)
- A: availability (dimensionless). MTBF and MTTR in seconds (s), minutes (min), or hours (h) — same time units must be used. Applies when MTBF and MTTR are defined as averages over an operation period.
Downtime per year (hours) = (1 - A) × 8760 h
- A: availability (dimensionless). 8760 h = hours in a non-leap year. Use to compute expected yearly downtime from availability SLA.
Note: For cloud architecture, remember important configuration syntax/concepts (no single syntax across providers):
- Infrastructure-as-Code examples: declare resources, providers, and dependencies (e.g., Terraform:
resource "aws_instance" "web" { ... }). Verify provider-specific syntax when coding.
Worked examples
Example 1 — Numeric availability using MTBF/MTTR
Given: MTBF = 1500 hours, MTTR = 5 hours. Find availability and expected downtime per year.
Steps:
- Write formula:
A = MTBF / (MTBF + MTTR). - Substitute: A = 1500 / (1500 + 5) = 1500 / 1505.
- Compute A = 0.99667774 ≈ 0.99668 (dimensionless) = 99.6678%.
- Downtime per year = (1 - A) × 8760 h = (1 - 0.99667774) × 8760 ≈ 0.00332226 × 8760 ≈ 29.10 hours.
Final answer: Availability = 99.6678% and expected downtime ≈ 29.10 hours/year.
Example 2 — Diagram-style model answer (short-essay exam question)
Question: Draw and explain a secure hybrid cloud architecture for a hospital storing patient records.
Model answer (what to draw and explain): Step 1 draw clients (doctors, admin) connecting over Internet and hospital LAN to a gateway. Step 2 draw a VPN/Direct Connect link to a private cloud where the primary electronic health record (EHR) database is hosted in a private subnet (label firewalls and IAM). Step 3 in the public cloud draw a compute cluster for web applications and analytics in separate subnets; show that sensitive PHI (Protected Health Information) remains in private cloud; non-sensitive workloads and analytics run in public cloud with encrypted transfer (TLS). Step 4 label backups and a separate disaster recovery site (multi-region replicas). Explain access control (multi-factor auth) and audit logging. Final sentence: this hybrid setup balances compliance (private) and scalability/cost (public).
Example 3 — Decision-making short problem
Question: A fintech startup needs rapid scale and low cost for customer-facing features but must keep customer transaction ledger within regulatory control. Which deployment model and architecture?
Answer: Use a hybrid cloud: public cloud for front-end, transaction processing microservices and caching (scale-out); private or dedicated cloud for the ledger database (to meet regulatory control). Use strong encryption for data in transit and at rest, a VPN or direct connect between public and private parts, and a broker for identity and audit. Justify by cost, performance, and compliance trade-offs.
Common mistakes
- Quoting CAP theorem as absolute (saying all three can never be achieved even without partitions) — must state assumption about partitions.
- Forgetting units when using MTBF/MTTR (mixing hours and minutes).
- Drawing architecture diagrams without labeling subnets, security boundaries, and data flow directions.
- Saying "private cloud is always more secure" — security depends on controls, not model alone.
For your exam
Typical question types:
- Definitions and short notes: define the four deployment models and list pros/cons.
- Diagram question: draw a hybrid cloud or N-tier cloud architecture and label components and data flow.
- Derivation/numerical: compute availability from MTBF/MTTR and compute downtime per year.
- Comparison/essay: argue for a deployment model for a given scenario (bank, e-commerce, research lab).
- Conceptual: explain CAP theorem and its implications for database choices in cloud.
What full-marks answers include:
- Precise definitions of terms (bolded in study notes) and assumptions.
- For diagrams: clear boxes, arrows showing data flow, labelled security boundaries (firewall, VPN), and notes on replication/failover.
- For derivations: write the formula, show algebraic steps, keep units consistent, give numeric results with units and a short conclusion.
- For comparison/essay: list requirements (security, cost, scalability, compliance), map them to pros/cons of each model, and conclude with a justified recommendation.
Quick check
- Define Hybrid cloud in one sentence.
- Write the formula relating Availability, MTBF and MTTR.
- State the three properties in the CAP theorem.
Answers: 1) A Hybrid cloud composes two or more distinct cloud infrastructures (public, private or community) with connectivity that allows data and application portability. 2) A = MTBF / (MTBF + MTTR). 3) Consistency, Availability, Partition tolerance.
Finished this topic? Mark it so your progress, study plan and readiness keep up.
Stuck on something here?