This free curriculum covers every domain assessed in the Oxford Certified Cloud Solutions Architect examination. It is vendor-neutral: the concepts apply to all major public cloud providers. Read each chapter carefully, note the key points, and then sit the 80-question examination.
Cloud computing delivers computing resources — servers, storage, databases, networking and software — on demand over the internet with pay-as-you-go pricing. The US National Institute of Standards and Technology (NIST) defines five essential characteristics: on-demand self-service, broad network access, resource pooling, rapid elasticity and measured service.
There are three main service models. In Infrastructure as a Service (IaaS) you rent virtual machines, storage and networks and manage the operating system and everything above it. In Platform as a Service (PaaS) the provider manages the operating system and runtime and you deploy your code. In Software as a Service (SaaS) you simply use and configure a finished application such as email or CRM.
Deployment models describe where the cloud runs: public cloud (shared provider infrastructure), private cloud (dedicated to one organisation), hybrid cloud (on-premises and public cloud working together) and multi-cloud (using several public providers).
IaaS gives the most control; SaaS gives the least control and the least management effort
Hybrid = on-premises + cloud managed together; multi-cloud = more than one public provider
Choose the model based on control needs, skills and speed
À retenir: The more the provider manages, the less you control — and the less you have to operate.
Chapitre 2
The shared responsibility model and well-architected principles
Security and operations in the cloud are shared. The provider is responsible for security "of" the cloud: physical data centres, hardware, the global network and the virtualisation layer. The customer is responsible for security "in" the cloud: data, identities, access rights, configuration, and — for IaaS — the guest operating system, its patches and the applications.
Cloud providers publish well-architected frameworks built on similar pillars: operational excellence, security, reliability, performance efficiency, cost optimisation and sustainability. An architect evaluates every design against these pillars and makes trade-offs consciously — for example, more redundancy improves reliability but increases cost.
Service level agreements (SLAs) describe the provider's availability commitments for each service, but the availability of your overall solution depends on how you combine services and remove single points of failure.
Your architecture, not only the provider's SLA, determines end-to-end availability
À retenir: Know exactly where the provider's responsibility ends and yours begins.
Chapitre 3
Compute: virtual machines, containers and serverless
Virtual machines provide full control of the operating system and suit traditional or legacy workloads. Containers package an application with its dependencies and share the host kernel, so they are lightweight, portable and start quickly. Orchestration platforms such as Kubernetes schedule containers across a cluster, restart failed containers and scale them automatically.
Serverless functions run code in response to events — an uploaded file, an API call, a message — without managing servers. You pay only for execution time, which makes serverless ideal for short, event-driven or unpredictable workloads.
Scaling can be vertical (a bigger instance) or horizontal (more instances). Horizontal scaling combined with auto scaling lets capacity follow demand. Pricing options include on-demand (flexible), reserved instances or savings plans (discount for commitment) and spot or preemptible capacity (very cheap but can be interrupted, suitable for fault-tolerant batch jobs).
VMs: maximum control; containers: portable and fast; serverless: no servers to manage
Kubernetes orchestrates, heals and scales containers
Horizontal scaling adds instances; vertical scaling enlarges one instance
Use auto scaling with schedules and metrics to match predictable and unpredictable demand
Spot capacity suits interruptible workloads only
À retenir: Pick the simplest compute model that meets the workload's needs.
Chapitre 4
Storage and databases
Object storage keeps files as objects in buckets accessed over HTTP; it is massively scalable and ideal for images, backups, logs and data lakes. Block storage provides disks attached to virtual machines, used for operating systems and databases. File storage offers shared file systems for several servers. Storage classes range from hot (frequent access) to cold and archive tiers, which are cheap but slower to retrieve.
Relational databases provide structured tables, SQL and strong transactions. NoSQL databases — key-value, document, wide-column and graph — offer flexible schemas and very high scale for specific access patterns. Managed database services automate backups, patching and high availability, reducing operational effort.
Read replicas offload read traffic from the primary database. In-memory caches reduce latency for frequently read data, but a cache is not durable storage. Backups and replication must be designed against two targets: the Recovery Point Objective (how much data you can afford to lose) and the Recovery Time Objective (how quickly service must be restored).
Object = files at scale; block = VM disks; file = shared file systems
Use lifecycle rules to move old data to cold or archive tiers
Relational for transactions; NoSQL for flexible, high-scale access patterns
RPO = acceptable data loss; RTO = acceptable downtime
À retenir: Match each data set to the storage and database type built for its access pattern.
Chapitre 5
Cloud networking
A virtual private cloud (VPC) or virtual network is a logically isolated network where you choose IP ranges, create subnets and control routing. Public subnets have a route to the internet gateway; private subnets do not. Databases and internal services belong in private subnets. A NAT gateway lets private instances make outbound connections (for example to download updates) without being reachable from the internet.
Security groups are stateful firewalls attached to instances; network access control lists are typically stateless and apply at subnet level. Load balancers distribute traffic: application (Layer 7) load balancers can route by URL path or host name, while network (Layer 4) load balancers handle very high-performance TCP/UDP traffic.
Content delivery networks cache content at edge locations close to users. Hybrid connectivity uses a site-to-site VPN (encrypted over the internet) or a dedicated private connection for consistent bandwidth and latency.
Place internet-facing components in public subnets and data stores in private subnets
VPN for encrypted connectivity; dedicated links for consistent large transfers
À retenir: Design the network so that only what must be public is public.
Chapitre 6
High availability and disaster recovery
A region is a geographic area containing several availability zones; each zone is one or more isolated data centres with independent power, cooling and networking. Deploying across multiple zones behind a load balancer protects against a data-centre failure. Multi-region designs protect against a regional outage but add complexity and cost.
Disaster-recovery strategies form a spectrum. Backup and restore is cheapest but slowest. Pilot light keeps a minimal core — usually replicated data — running in the recovery region and scales it up when needed. Warm standby runs a scaled-down but functional copy. Active-active runs full production in several regions at once for the lowest RTO and RPO and the highest cost.
Remove single points of failure: run multiple instances, use managed databases with standby replicas, configure health checks so load balancers stop sending traffic to unhealthy instances, and test failover regularly.
Multi-AZ for data-centre failures; multi-region for regional failures
DR spectrum: backup & restore → pilot light → warm standby → active-active
Lower RTO/RPO costs more — choose according to business impact
Health checks and automatic failover keep services available
Test recovery plans; an untested backup is not a backup
À retenir: Design for failure: assume any component can fail and plan how the system survives it.
Chapitre 7
Identity, access and data protection
Identity and access management (IAM) controls who can do what. Apply least privilege: grant only the permissions required and review them regularly. Applications should use roles with temporary credentials rather than long-lived access keys stored in code. Protect the root or owner account with multi-factor authentication, do not use it for daily work and never create shared keys for it.
Encrypt data at rest and in transit (TLS). A key management service creates, stores, rotates and controls access to encryption keys. Secrets — database passwords, API keys — belong in a secrets manager with rotation, never in source code, even in private repositories.
Enable audit logging of all API calls to record who did what and when. Block public access to storage by default; if a bucket is found public by mistake, remove the access immediately, review the logs and enforce a preventive policy. The zero-trust model verifies identity, device and context for every request instead of trusting the network location.
Least privilege, roles instead of keys, MFA everywhere
Encrypt at rest (KMS) and in transit (TLS)
Secrets manager for credentials, with rotation
Audit logs for every account; block public storage access by default
Zero trust: never trust by default, always verify
À retenir: Most cloud breaches come from misconfiguration and over-permissive access — not from the provider.
Chapitre 8
Modern application architecture
A monolith is one deployable unit; microservices split an application into small services that can be developed, deployed and scaled independently, at the cost of more operational complexity. Stateless application tiers store session data outside the instance (for example in a cache or database), so any instance can serve any request and the tier scales horizontally.
Loose coupling improves resilience. Message queues hold messages until consumers can process them, so a slow component no longer blocks others. Publish/subscribe topics deliver one event to many subscribers. Event-driven architectures react to events such as "order created" and make it easy to add new consumers.
An API gateway sits in front of services to handle authentication, rate limiting, routing, request transformation and monitoring.
Microservices: independent deployment and scaling, more complexity
Stateless tiers scale horizontally
Queues decouple producers and consumers; topics fan out events
Do slow, non-critical work asynchronously (for example confirmation emails)
API gateways centralise authentication, throttling and routing
Cloud migrations are planned with the "6 Rs": rehost (lift and shift with little change), replatform (lift, tinker and shift — small optimisations such as moving to a managed database), repurchase (move to a SaaS product), refactor or re-architect (rebuild using cloud-native services), retire (switch off what is no longer needed) and retain (keep on-premises for now).
A migration starts with discovery and assessment: inventory applications, dependencies, performance and licences. A total cost of ownership (TCO) comparison includes infrastructure, licences, staff effort, facilities and the migration itself — not just server prices. Applications are then grouped into waves and moved in phases, starting with lower-risk workloads, with rollback plans for each wave.
Rehost is fastest; refactor delivers the most cloud benefit but takes longest
Repurchase = replace with SaaS; retire = switch off unused systems
Assess dependencies before moving anything
Compare full TCO, not only compute prices
Migrate in waves with clear rollback plans
À retenir: Choose a migration strategy per application, based on business value and effort.
Chapitre 10
Cost optimisation and FinOps
Cloud costs are variable, so they must be managed continuously. FinOps brings finance, technology and business together to take ownership of cloud spending. Core practices include tagging every resource by owner, project and environment; setting budgets and alerts; reviewing idle resources; and showing teams their own costs.
Rightsizing matches instance sizes to real usage. Commitment discounts (reserved instances or savings plans) reduce the price of steady workloads; spot capacity reduces the price of interruptible ones; schedules stop non-production environments outside working hours. Data transfer (egress) charges are often overlooked and should be considered in architecture decisions. Sustainable designs — managed services, serverless and rightsizing — also reduce idle capacity and energy use.
Tag resources so every cost has an owner
Budgets and alerts catch surprises early
Rightsize, schedule and delete unused resources
Use commitments for steady load and spot for interruptible load
Remember data-transfer costs
À retenir: In the cloud you pay for what you provision — so provision only what you need.
Chapitre 11
Cloud operations and observability
Infrastructure as Code (IaC) defines environments in version-controlled templates so they can be created consistently and repeatedly. CI/CD pipelines build, test and deploy changes automatically. Immutable infrastructure replaces servers with new ones built from updated images instead of changing them in place.
Safe release techniques reduce risk. Blue/green deployments keep two identical environments and switch traffic, allowing instant rollback. Canary releases expose a new version to a small group first; if errors rise, the rollout is stopped and rolled back. Observability combines metrics, logs and traces to understand system behaviour. Alerts should be actionable — each one should require a human response.
Compliance requirements such as data residency are met by choosing the right regions and restricting replication to permitted locations.
IaC + CI/CD = repeatable, reviewable change
Blue/green and canary releases make rollback easy
Metrics, logs and traces are the three pillars of observability
Only create alerts someone must act on
Respect data-residency rules when choosing regions
À retenir: Automate everything you can, observe everything that matters, and always be able to roll back.
Prêt ? Passez l’examen gratuit
Passez l’examen et obtenez gratuitement votre badge numérique. Après la réussite, vous pourrez commander le certificat officiel.