CLOUD COMPUTING

Comprehensive Study Notes

Architecture · Services · Deployment · Configuration
1. Components of Cloud Architecture
Cloud architecture defines how the technology components of cloud computing are
interconnected to provide scalable, reliable, and on-demand computing services.

1.1 Regions
A region is a geographic area containing multiple data centres. Cloud providers deploy regions
worldwide to allow customers to run workloads close to their users.
▸ Regions are completely independent of each other to ensure fault isolation
▸ Customers choose a region based on latency, compliance, and data residency
requirements
▸ Examples: AWS us-east-1 (N. Virginia), Azure East US, GCP europe-west1

1.2 Availability Zones (AZs)

Availability Zones are physically separate data centres within a region, connected via low-
latency private networks.
▸ Each AZ has independent power, cooling, and networking
▸ Running resources across multiple AZs provides high availability
▸ Failure in one AZ does not affect other AZs in the same region
▸ Typically 2–6 AZs per region

AZ vs Region

A Region contains multiple Availability Zones. AZs protect against data centre failures, while
multiple Regions protect against large-scale regional disasters.

1.3 Data Centres

Physical facilities housing computing, storage, and networking hardware. They form the
foundational infrastructure of cloud services.
▸ Feature redundant power supplies, UPS systems, and diesel generators
▸ Use precision air cooling or liquid cooling systems
▸ Physical security includes biometric access, 24/7 surveillance, and security guards
▸ Interconnected with high-bandwidth fibre networks

1.4 Virtualisation
Virtualisation is the creation of virtual (rather than physical) versions of computing resources,
including servers, storage, and networks.
▸ Enables multiple virtual machines to run on a single physical server
▸ Improves hardware utilisation rates from ~10% to 60–80%
▸ Isolates workloads for security and stability
▸ Supports rapid provisioning and decommissioning of resources

1.5 Virtual Machines (VMs)

A Virtual Machine is a software-based emulation of a physical computer. VMs run a full OS
instance and are isolated from each other.

Property Description
Guest OS Each VM runs its own independent operating system
Isolation VMs are fully isolated — one VM crash does not affect others
Portability VMs can be moved between physical hosts (live migration)
Snapshots VM state can be saved and restored at any point in time
Resource Allocation CPU, RAM, and storage are allocated from the physical host pool

1.6 Hypervisor Types

A hypervisor (Virtual Machine Monitor) creates and manages VMs by abstracting hardware
resources.

Type 1 — Bare Metal Hypervisor

Runs directly on the physical hardware without a host operating system.
▸ Direct access to hardware — higher performance and efficiency
▸ Used in enterprise and cloud environments
▸ Examples: VMware ESXi, Microsoft Hyper-V, Xen, KVM (Linux kernel-based)

Type 2 — Hosted Hypervisor

Runs on top of an existing host operating system.
▸ Easier to set up — good for development and testing
▸ More overhead due to host OS layer
▸ Examples: VMware Workstation, Oracle VirtualBox, Parallels

Hypervisor Type Characteristics

 Type 1 Bare metal, runs on hardware directly, high performance, used in
production cloud
Type 2 Runs on host OS, easier setup, lower performance, used for
development/testing

1.7 Networking in Cloud Architecture

Cloud networking provides connectivity between cloud resources and to the internet.
▸ Virtual Private Cloud (VPC): isolated network environment within the cloud
▸ Subnets: subdivisions of a VPC for organising resources
▸ Security Groups / Network ACLs: control inbound and outbound traffic
▸ Load Balancers: distribute traffic across multiple instances
▸ CDN (Content Delivery Network): caches content at edge locations globally
▸ VPN / Direct Connect: secure private connections to on-premises networks
▸ Software-Defined Networking (SDN): programmatic control of network infrastructure

1.8 Types of Cloud Storage

Storage Type Description Examples
Block Storage Low-level storage divided into AWS EBS, Azure Disk, Google
blocks. Attached to VMs like a Persistent Disk
hard drive.
Object Storage Stores data as objects with AWS S3, Azure Blob, Google
metadata and unique IDs. Highly Cloud Storage
scalable.
File Storage Shared file system accessible over AWS EFS, Azure Files, Google
standard protocols (NFS, SMB). Filestore
Archive Storage Low-cost, slow-access storage for AWS Glacier, Azure Archive,
long-term retention. Google Coldline
In-Memory Ultra-fast storage held in RAM for Redis, Memcached, Amazon
caching and real-time processing. ElastiCache
2. Application Migration Approaches
Application migration is the process of moving software applications from one computing
environment to another. The approach chosen depends on business goals, budget, timeline,
and technical constraints.

The 7 R's of Migration

The most widely used framework for cloud migration strategies: Retire, Retain, Rehost,
Replatform, Repurchase, Refactor/Re-architect, and Relocate.

2.1 Rehost — 'Lift and Shift'

Move applications as-is from on-premises to the cloud with no code changes.
▸ Fastest migration approach with minimal risk
▸ No optimisation for cloud-native features
▸ Good for legacy applications where refactoring is costly
▸ Cost savings come from eliminating on-premises hardware
▸ Tools: AWS Server Migration Service, Azure Migrate, Google Migrate

2.2 Replatform — 'Lift, Tinker and Shift'

Move applications with minor optimisations to take advantage of cloud capabilities without
changing core architecture.
▸ Example: migrating a database to a managed service (RDS, Cloud SQL)
▸ Reduces operational overhead while preserving existing code
▸ Moderate effort — some code or configuration changes required
▸ Better performance and reduced admin compared to pure rehost

2.3 Refactor / Re-architect

Redesign the application to be cloud-native, often adopting microservices, containers, or
serverless architectures.
▸ Highest effort and cost but delivers greatest long-term benefits
▸ Enables full use of cloud-native features: auto-scaling, managed services, serverless
▸ Breaks monolithic applications into independent microservices
▸ Ideal when business requirements demand significant performance or agility
improvements
2.4 Repurchase — 'Drop and Shop'
Replace the existing application with a commercial SaaS product.
▸ Example: replacing an on-premises CRM with Salesforce
▸ Eliminates maintenance burden of running the software
▸ May involve data migration and user retraining
▸ Best when a better SaaS alternative exists for the business function

2.5 Retain — 'Revisit'

Keep certain applications on-premises because migration is not yet justified.
▸ Applicable to recently upgraded systems or applications with compliance requirements
▸ May be revisited in a future migration wave
▸ Hybrid approach — some systems in cloud, others on-premises

2.6 Retire
Decommission applications that are no longer needed.
▸ Reduces total portfolio size and cost
▸ Eliminates security risks from unused systems
▸ Identifies redundant applications during the discovery phase

2.7 Relocate
Move infrastructure to the cloud without purchasing new hardware or re-writing applications.
Distinct from rehost as it moves entire virtualised environments.
▸ Example: moving VMware workloads to VMware Cloud on AWS
▸ Minimal disruption to operations
▸ Retains existing tools and processes

Strategy Summary
Rehost No changes. Fastest. Least cloud benefit.
Replatform Minor optimisations. Moderate effort. Better efficiency.
Refactor Full redesign. Highest effort. Maximum cloud benefit.
Repurchase Replace with SaaS. Low IT effort. Vendor dependency.
Retain Keep on-premises. No migration effort. Deferred decision.
Retire Decommission. Reduces cost and complexity.
Relocate Move virtualised env. Minimal disruption. Same tooling.
3. Key Characteristics of Cloud Computing
NIST defines five essential characteristics of cloud computing, with additional characteristics
recognised in practice.

3.1 On-Demand Self-Service

Users can provision computing resources automatically without requiring human interaction with
the service provider.
▸ Resources available through a web portal or API at any time
▸ No need to contact a salesperson or wait for procurement
▸ Enables rapid experimentation and agile development

3.2 Broad Network Access

Services are available over the network and accessible through standard mechanisms from
diverse devices.
▸ Accessible from laptops, tablets, smartphones, and workstations
▸ Uses standard protocols (HTTPS, REST APIs)
▸ Enables device and location independence — work from anywhere

3.3 Resource Pooling

Computing resources are pooled to serve multiple consumers using a multi-tenant model, with
resources dynamically assigned and reassigned.
▸ Physical and virtual resources shared across customers
▸ Customers typically do not know the exact location of their resources
▸ Enables economies of scale for the provider

3.4 Rapid Elasticity

Resources can be elastically provisioned and released, in some cases automatically, to scale
rapidly with demand.
▸ Scale out: add more instances during peak load
▸ Scale in: reduce instances during low demand to save cost
▸ Auto-scaling policies respond to CPU, memory, or queue depth metrics
▸ Appears to the consumer as unlimited available capacity

3.5 Measured Service

Cloud systems automatically control and optimise resource use by metering capabilities. Usage
is monitored, controlled, and reported.
▸ Pay only for what you use (pay-per-use model)
▸ Granular billing: per-second, per-request, per-GB
▸ Detailed usage reports for cost management and chargebacks

3.6 Device and Location Independence

Users can access services from any device at any location without being tied to specific
hardware or a physical office.
▸ Data and applications live in the cloud, not on a local device
▸ Enables remote work, BYOD (Bring Your Own Device) policies
▸ Reduces dependency on corporate-owned hardware

3.7 Multi-Tenancy and Shared Resources

Multiple customers (tenants) share the same underlying physical infrastructure while their data
and workloads remain logically isolated.
▸ Isolation achieved through virtualisation and logical separation
▸ Cost efficiency — hardware costs shared across many tenants
▸ Security controls prevent one tenant from accessing another's data

3.8 Centralised Performance Monitoring

Cloud providers and customers can monitor the health, performance, and availability of cloud
resources from a central dashboard.
▸ Real-time metrics: CPU, memory, disk I/O, network throughput
▸ Alerting and automated responses to threshold breaches
▸ Services: AWS CloudWatch, Azure Monitor, Google Cloud Monitoring

3.9 Hardware Independence and Variable Cost

Applications run without being tied to specific physical hardware. Costs vary with usage rather
than being fixed capital expenses.
▸ Shift from CapEx (capital expenditure) to OpEx (operational expenditure)
▸ No upfront hardware purchases — pay as you go
▸ Hardware upgrades handled transparently by the provider

3.10 Distribution Over the Internet

Cloud services are delivered over public or private internet connections, making them globally
accessible.
▸ Content Delivery Networks (CDNs) cache data at edge locations worldwide
▸ Reduces latency for geographically distributed users
▸ Enables global application deployment without physical offices

3.11 Improved Business Continuity

Cloud infrastructure supports disaster recovery and high availability strategies that would be
cost-prohibitive on-premises.
▸ Data replicated across multiple Availability Zones and Regions
▸ Automated backups and point-in-time recovery
▸ Recovery Time Objective (RTO) and Recovery Point Objective (RPO) can be
dramatically reduced
▸ Managed failover ensures minimal downtime during outages
4. Cloud Computing Service Models
Cloud services are delivered in distinct models that determine the level of control, flexibility, and
management responsibility between the provider and the customer.

4.1 Infrastructure as a Service (IaaS)

The provider delivers virtualised computing infrastructure over the internet. The customer
manages the OS, middleware, runtime, and applications.
▸ Provides: virtual machines, storage, networking, firewalls
▸ Customer manages: OS, middleware, applications, data
▸ Provider manages: physical hardware, hypervisors, data centre
▸ Use cases: hosting websites, running custom enterprise applications, development/test
environments
▸ Examples: AWS EC2, Azure Virtual Machines, Google Compute Engine

4.2 Platform as a Service (PaaS)

The provider delivers a platform allowing customers to develop, run, and manage applications
without managing underlying infrastructure.
▸ Provides: runtime environment, databases, middleware, development tools
▸ Customer manages: applications and data only
▸ Provider manages: OS, servers, storage, networking
▸ Use cases: web application development, API backends, data analytics
▸ Examples: AWS Elastic Beanstalk, Azure App Service, Google App Engine, Heroku

4.3 Software as a Service (SaaS)

Fully managed software applications delivered over the internet. The customer only uses the
software — nothing else is managed.
▸ Provides: complete application accessible via web browser or API
▸ Customer manages: user accounts and data input only
▸ Provider manages: everything — infrastructure, platform, application
▸ Use cases: email, CRM, collaboration tools, HR systems
▸ Examples: Gmail, Microsoft 365, Salesforce, Dropbox, Zoom, SAP

4.4 Mobile Backend as a Service (MBaaS)

A cloud service model that provides backend services specifically designed for mobile and web
applications.
 ▸ Provides: user authentication, push notifications, cloud storage, database, social
integrations
▸ Allows developers to focus on frontend/mobile app development
▸ Eliminates need to build and maintain backend infrastructure
▸ Examples: Firebase (Google), AWS Amplify, Back4App, Supabase

4.5 Serverless Computing

A cloud execution model where the cloud provider dynamically manages infrastructure
allocation. Developers write code; the provider handles everything else.
▸ No server provisioning, patching, or capacity planning required
▸ Applications run in response to events (HTTP requests, file uploads, timers)
▸ Automatically scales to zero when not in use — no cost when idle
▸ Billing based on number of executions and execution time
▸ Best for event-driven, stateless workloads

4.6 Function as a Service (FaaS)

A subset of serverless computing where individual functions are deployed and executed in
response to events.
▸ Finest granularity of serverless — deploy individual functions
▸ Stateless by design — each invocation is independent
▸ Supports multiple triggers: HTTP, queue messages, database changes, schedules
▸ Cold start latency is a consideration for latency-sensitive applications
▸ Examples: AWS Lambda, Azure Functions, Google Cloud Functions

Model What's Provided Key Characteristics

IaaS VMs, storage, networking Full control, high responsibility
PaaS Platform + runtime Moderate control, less admin
SaaS Complete application No control, zero admin
MBaaS Mobile/web backend Mobile-focused, rapid
development
Serverless Auto-managed compute Event-driven, no server
management
FaaS Individual functions Maximum granularity, per-
execution billing
5. Techniques and Methods for Cloud Deployment

5.1 Networking
Cloud networking enables secure, scalable communication between resources.
▸ Virtual Private Cloud (VPC): isolated network for your cloud resources
▸ Peering: connect two VPCs or on-premises networks privately
▸ Transit Gateway: hub-and-spoke model connecting many VPCs
▸ Private Link / Service Endpoints: access cloud services without traversing the public
internet
▸ DNS Management: Route 53 (AWS), Azure DNS, Cloud DNS for service discovery
▸ BGP Routing: dynamic routing for hybrid connectivity

5.2 Automation and Self-Service

Infrastructure automation removes manual processes and enables consistent, repeatable
deployments.
▸ Infrastructure as Code (IaC): define infrastructure in code files — Terraform, AWS
CloudFormation, Azure Bicep
▸ CI/CD Pipelines: automate build, test, and deployment — GitHub Actions, Jenkins,
Azure DevOps
▸ Configuration Management: enforce desired state — Ansible, Chef, Puppet
▸ Self-service portals: allow teams to provision approved resources independently
▸ GitOps: treat infrastructure changes like code changes with pull requests and review

5.3 Federation
Federation enables identity and access management across multiple cloud providers or
between cloud and on-premises systems.
▸ Single Sign-On (SSO): users authenticate once and access multiple systems
▸ SAML 2.0 / OAuth 2.0 / OpenID Connect: standard protocols for federated identity
▸ Active Directory Federation Services (ADFS): extends on-premises AD to cloud
▸ Cloud-to-cloud federation: unified identity across AWS, Azure, and GCP
▸ Enables hybrid cloud scenarios where users access both cloud and on-premises
resources

5.4 The Role of Standardisation

Standards ensure interoperability, portability, and security across cloud environments.
▸ Open Container Initiative (OCI): container image and runtime standards
▸ Kubernetes: de facto standard for container orchestration
▸ OpenAPI / REST: standard API design for cloud services
▸ ISO/IEC 27001, SOC 2, PCI-DSS: security and compliance standards
▸ TOSCA (Topology and Orchestration Specification for Cloud Applications): cloud
portability standard
▸ Standardisation reduces vendor lock-in and enables multi-cloud strategies
6. Cloud Configuration

6.1 Cloud Service Models (Recap and Responsibility)

Service model responsibility defines who manages what between the cloud provider and the
customer.

Responsibility IaaS PaaS / SaaS

Physical Hardware Provider Provider
Hypervisor / Provider Provider
Virtualisation
Operating System Customer Provider
Middleware / Runtime Customer Provider
Application Code Customer Customer (SaaS: Provider)
Data Customer Customer
Identity & Access Shared Shared

6.2 Derivative Service Models

Beyond the core IaaS/PaaS/SaaS models, several derivative models have emerged:
▸ DaaS — Desktop as a Service: virtual desktops delivered via cloud (Citrix, AWS
WorkSpaces)
▸ DBaaS — Database as a Service: managed database engines (RDS, Cosmos DB,
Cloud SQL)
▸ STaaS — Storage as a Service: managed cloud storage (S3, Azure Blob)
▸ SECaaS — Security as a Service: cloud-delivered security (Cloudflare, Zscaler)
▸ AIaaS — AI as a Service: pre-built AI/ML models via API (AWS Rekognition, Azure
Cognitive Services)
▸ NaaS — Network as a Service: on-demand networking resources

6.3 Delivery Models

Cloud deployment can follow four primary delivery models:

Model Description
 Public Cloud Infrastructure owned and operated by a third-party provider,
shared among customers. Examples: AWS, Azure, GCP.
Private Cloud Infrastructure operated exclusively for a single organisation, on-
premises or in a dedicated facility. More control and security.
Hybrid Cloud Combines public and private cloud, allowing data and applications
to move between them. Balances flexibility and control.
Multi-Cloud Uses services from two or more public cloud providers. Avoids
vendor lock-in and optimises costs by choosing best services per
provider.
Community Cloud Shared infrastructure for a specific community with shared
concerns (e.g., government agencies, healthcare organisations).

6.4 On-Premises vs Off-Premises Hosting

On-Premises: Infrastructure is located within the organisation's own facilities.
▸ Full control over hardware, software, and data
▸ High capital expenditure (CapEx) for hardware
▸ Responsible for all maintenance, upgrades, and security
▸ Required for certain regulated industries with strict data residency rules

Off-Premises (Cloud Hosted): Infrastructure managed by a cloud provider in remote data

centres.
▸ Operational expenditure (OpEx) model — pay as you go
▸ Provider handles hardware maintenance and upgrades
▸ Faster deployment, global reach
▸ Data sovereignty considerations must be addressed

6.5 Orchestration Platforms

Orchestration automates the deployment, scaling, and management of containerised
applications and cloud resources.
▸ Kubernetes (K8s): leading container orchestration platform. Automates deployment,
scaling, and self-healing of containers.
▸ Docker Swarm: simpler container clustering, built into Docker
▸ Apache Mesos: distributed systems kernel for resource management
▸ Managed K8s: AWS EKS, Azure AKS, Google GKE — reduce operational overhead
▸ Terraform / Pulumi: orchestrate infrastructure provisioning across multiple clouds
▸ Helm: package manager for Kubernetes applications
6.6 Cloud Bursting
Cloud bursting is a deployment model where an application runs in a private cloud or on-
premises and bursts into a public cloud when computing capacity demand increases.
▸ Baseline workload handled by private cloud / on-premises infrastructure
▸ Peak demand overflow automatically directed to public cloud resources
▸ Optimises costs — public cloud used only when needed
▸ Requires compatible networking between on-premises and cloud (VPN or Direct
Connect)
▸ Applications must be designed to run in both environments
▸ Examples: running HPC simulations or seasonal e-commerce traffic spikes

Cloud Bursting Use Case

A retailer handles normal daily traffic on-premises but automatically bursts to AWS during Black
Friday, paying only for the extra capacity used during the peak period.

6.7 Cloud Terms and Characteristics

Term Definition
Scalability Ability to increase or decrease resources to handle changing
workloads
Elasticity Automatic and rapid scaling in response to real-time demand
changes
High Availability (HA) System design ensuring minimal downtime through redundancy
Fault Tolerance Ability to continue operating despite component failures
Disaster Recovery (DR) Processes to restore services after a major outage or disaster
SLA Service Level Agreement — contractual uptime and performance
guarantees
Latency Time taken for a request to travel from client to server and back
Throughput Amount of data successfully processed in a given time period
Provisioning Process of setting up IT infrastructure resources
Deprovisioning Process of removing and releasing cloud resources no longer
needed
Tenant A customer using shared cloud infrastructure
Instance A single virtual machine running in the cloud
Workload Any application, service, or process running in the cloud
6.8 Storage Concepts
Cloud storage underpins almost all cloud workloads. Understanding storage concepts is
essential for architecture decisions.

Object Storage
Stores data as objects (files + metadata + unique ID). Infinitely scalable, no folder hierarchy.
▸ Accessed via REST APIs (GET, PUT, DELETE)
▸ Ideal for unstructured data: images, videos, backups, logs
▸ No size limit on total storage — petabyte scale
▸ Features: versioning, lifecycle policies, cross-region replication

Block Storage
Raw storage divided into fixed-size blocks, presented as a virtual disk to the OS.
▸ Low-latency, high-IOPS performance for databases and OS volumes
▸ Attached to a single VM at a time (or shared block with special configuration)
▸ Supports file systems: NTFS, ext4, XFS

File Storage (Shared File Systems)

Hierarchical file system accessible by multiple VMs simultaneously.
▸ NFS (Linux) or SMB/CIFS (Windows) protocols
▸ Ideal for shared content, home directories, CMS platforms
▸ Higher latency than block storage

Storage Tiers
Tier Characteristics
Hot / Frequent Access High-performance, higher cost. Data accessed multiple times per
month.
Cool / Infrequent Access Lower cost, retrieval fees. Data accessed a few times per year.
Archive / Cold Very low cost, high retrieval time (hours). Long-term retention.
Intelligent Tiering Automatically moves data between tiers based on access patterns.

Key Storage Concepts

▸ IOPS (Input/Output Operations Per Second): measure of storage performance
▸ Throughput: data transfer rate (MB/s)
▸ Replication: data copied to multiple locations for durability (e.g., 99.999999999% / 11
nines)
▸ Snapshots: point-in-time copies of volumes for backup
▸ Encryption at rest: data encrypted when stored; at transit: encrypted in network
▸ Data lifecycle policies: automate moving/deleting data based on age or access
▸ Consistency models: strong consistency (immediate reads reflect writes) vs eventual
consistency
7. Quick Reference Summary

Acronym/Term Full Name Key Point

Rehost Lift & Shift No change, fastest migration
Replatform Lift, Tinker & Shift Minor cloud optimisations
Refactor Re-architect Full cloud-native redesign
IaaS Infra as a Service VMs, storage, networking
PaaS Platform as a Service Managed runtime + DB
SaaS Software as a Service Full app, no management
FaaS Function as a Service Event-driven functions
MBaaS Mobile Backend as a Service Backend for mobile apps
AZ Availability Zone Isolated DC within a region
VPC Virtual Private Cloud Isolated cloud network
CDN Content Delivery Network Edge caching for low latency
HA High Availability Redundancy to minimise
downtime
DR Disaster Recovery Restore after major failure
IaC Infrastructure as Code Define infra in version-controlled
files
K8s Kubernetes Container orchestration platform

End of Cloud Computing Notes