| Cloud computing | On-demand computing resources over the internet, paid by use | What the cloud is |
| Five characteristics | On-demand self-service, broad network access, pooling, rapid elasticity, measured service | What the cloud is |
| IaaS, PaaS, SaaS; public, private, hybrid, community | NIST's service and deployment models | What the cloud is |
| Hybrid / multi-cloud | Own and rent together / more than one provider | What the cloud is |
| Elasticity / scalability | Follows demand down as well as up / can grow | Elasticity |
| Scale up / scale out | A bigger machine / more machines | Elasticity |
| CapEx → OpEx | A purchase becomes a monthly bill | Capital and operating expense |
| Multi-tenancy | Shared hardware, separate machines | Capital and operating expense |
| Six advantages (AWS) | Trade CapEx for variable expense, economies of scale, stop guessing capacity, speed, no data centres, global in minutes | Agility |
| Egress | Data in is free; data out is billed | When the cloud costs more |
| AWS, Azure, Google Cloud | About 63% of the market together | The providers |
| VM / object storage / managed SQL | EC2, Azure VMs, Compute Engine / S3, Blob, Cloud Storage / RDS, Azure SQL, Cloud SQL | The providers |
| IaaS | You run the OS and up: EC2, Azure VMs, Compute Engine | IaaS: infrastructure as a service |
| PaaS | You bring code and data: Elastic Beanstalk, App Service, App Engine | PaaS: platform as a service |
| SaaS | You use it and own your data and users: Microsoft 365, Google Workspace | SaaS: software as a service |
| Shared responsibility | Provider: security of the cloud. You: security in the cloud | Shared responsibility |
| Always the provider's | Physical data centres, hosts and network | Shared responsibility by model |
| Inherited, shared, customer-specific | Control types in the AWS model | Shared responsibility by model |
| Always yours | Data, identities and accounts, devices | What is always yours |
| Open bucket | Not a hack: a customer setting (Capital One, 2019) | What is always yours |
| Compliance | A certified cloud covers the floor, not what you build on it | Compliance |
| Rule of thumb | Take the most managed option that fits | Choosing a service model |
| Region | A geographic area you choose; resources stay there | Regions |
| Availability zone | One or more data centres with own power and network; a region has 3 or more | Availability zones |
us-east-1a / eastus zone 1 / us-central1-a | Region and zone names on AWS, Azure, Google Cloud | Availability zones |
| Single-zone vs multi-zone SLA | About 99.5% for one machine / 99.99% across zones | Why zones exist |
| Latency floor | About 1 ms per 100 km, there and back | Latency |
| Choosing a region | Law, distance to users, services offered, price | Choosing a region |
| Region pairs (Azure) | Updates rolled one region at a time; recovery priority | Choosing a region |
| Zonal, regional, global | Disk / most services / IAM, DNS, CDN | Resource scope |
| Edge location | Small site for CDN and DNS: CloudFront, Front Door, Cloud CDN | Edge locations |
| VM, container, function, platform | Four ways to run code, most to least you manage | Compute |
m5.large | Family m, generation 5, size large | Virtual machines |
| General, compute, memory, storage, accelerated | Machine families by workload | Virtual machines |
| On-demand, reserved/savings plan, spot, dedicated | Ways to pay for the same machine | Virtual machines |
| Image, template, group | Auto Scaling group / scale set / managed instance group | Autoscaling |
| Container image → registry → run | ECR, ACR, Artifact Registry | Containers |
| EKS / AKS / GKE | Managed Kubernetes; ECS and Cloud Run as simpler options | Container orchestration |
| Lambda / Azure Functions / Cloud Run functions | Serverless functions: event driven, billed per ms | Serverless functions |
| Lambda limits | 15 minutes, up to 10 GB memory; cold start on first call | Serverless functions |
| Rule of thumb | Take the least you have to manage | Choosing compute |
| Object / block / file | Whole files by name / a disk for one machine / a shared folder | Storage |
| S3, Blob Storage, Cloud Storage | Object storage on each provider | Object storage |
| 11 nines | 99.999999999% durability: copies across three zones | Durability and availability |
| LRS, ZRS, GRS, GZRS | Azure: one data centre, three zones, two regions, both | Durability and availability |
| Standard, Infrequent Access, Glacier | Pay less to store, more to read | Storage classes |
| Lifecycle rule | Move or delete objects as they age | Storage classes |
| EBS / Managed Disks / Persistent Disk | Block storage on each provider | Block storage |
| Snapshot | Copies only blocks changed since the last one | Block storage |
| EFS / Azure Files / Filestore | File storage on each provider | File storage |
| Example prices (us-east-1, 2025) | S3 $0.023, EBS gp3 $0.08, EFS $0.30 per GB-month | Choosing storage |
| Snowball / Data Box | Ship a device when the wire is too slow | Data transfer |
| VPC / VNet / VPC | Your private network on AWS, Azure, Google Cloud | Virtual networks |
| 10.0.0.0/16 | 65,536 addresses; /24 subnets of 256 | Virtual networks |
| Public subnet | Its route table has a route to an internet gateway | Public and private subnets |
| NAT gateway | Out to the internet, nothing in | Public and private subnets |
| Security group / network ACL | Stateful, per instance / stateless, per subnet | Security groups |
| Shield, WAF | DDoS protection; filter web attacks | Security groups |
| ALB / NLB | Layer 7, by path and host / layer 4, by port | Load balancers |
| Route 53, Azure DNS, Cloud DNS | DNS; weighted, latency, failover, geolocation routing | DNS |
| VPN / Direct Connect, ExpressRoute, Interconnect | Over the internet / a private line | Connecting your own building |
| Peering / transit hub | Two networks / many networks | Connecting your own building |
| Private endpoint | Reach a service without the internet | Connecting your own building |
| Data transfer (2025) | Out $0.09/GB, between zones $0.01/GB each way, NAT $0.045/GB | The whole path |
| Managed database | Provider patches, backs up and fails over; you keep schema and queries | Managed databases |
| Multi-AZ / read replica | A standby for failover (60 to 120 s) / copies that take reads | Managed databases |
| Point-in-time restore | Backups kept 1 to 35 days, restore to the minute | Managed databases |
| RDS, Aurora / Azure SQL / Cloud SQL, Spanner | Relational services | Relational databases |
| DynamoDB / Cosmos DB / Firestore, Bigtable | NoSQL services | NoSQL databases |
| Key-value, document, wide-column, graph | The NoSQL family | NoSQL databases |
| Eventual vs strong consistency | Fast, maybe slightly old / always current | NoSQL databases |
| Redshift / Synapse, Fabric / BigQuery | Data warehouses | Data warehouses |
| ElastiCache / Azure Cache for Redis / Memorystore | Caches | Caches |
| DMS / Database Migration Service | Move an existing database to the cloud | Caches |
| SQS, SNS / Service Bus / Pub/Sub | Queues and topics | Queues |
| Authentication / authorisation | Who are you / may you do this | Identity and access management |
| Evaluation | Denied by default; an explicit Deny always wins | Identity and access management |
| User, group, role | Attach permissions to groups and roles, not people | Users, groups and roles |
| Root user | Lock away with MFA; never use day to day | Users, groups and roles |
| Policy | Effect, action, resource, condition | Policies |
| Azure RBAC / Google IAM | A role, assigned to someone, at a scope | Policies |
| Least privilege | Only what the job needs | Least privilege |
| Defence in depth / zero trust | Many layers / verify every request | Least privilege |
| MFA | Blocks over 99.9% of account attacks (Microsoft, 2019) | Multi-factor authentication |
| IAM Identity Center / Entra ID | Single sign-on to every account and app | Multi-factor authentication |
| Instance role / managed identity / service account | Temporary credentials for software, no stored keys | Identities for software |
| KMS, Key Vault, Secrets Manager | Keys and secrets kept out of code | Identities for software |
| CloudTrail / Activity Log / Cloud Audit Logs | Who did what, and when | Accounts and organisations |
| Three meters | Time switched on, space used, data sent out | The cloud bill |
| Waste | Idle machines, unattached disks, old snapshots, idle IPs | The cloud bill |
| Consolidated billing | One bill for many accounts, volume discounts shared | The cloud bill |
| On-demand / committed / spot | Full price / up to about 72% off for 1 or 3 years / up to 90% off, can be taken back | Pricing models |
| Savings Plans, Reservations, CUDs | Commitment discounts by provider | Pricing models |
| Right-sizing | Match the machine to its real use | Idle resources |
| Budgets and alerts | Warn during the month, not on the invoice | Budgets and alerts |
| Cost Explorer / Cost Management / Billing reports | See where the money goes | Budgets and alerts |
| Tags | Labels that split one bill by team or project | Tags |
| TCO | Compare everything owning costs, not the server price | Total cost of ownership |
| FinOps | Inform, optimise, operate, in a loop | Managing cost |
| 99.9% / 99.99% / 99.999% | About 8 h 46 min / 53 min / 5 min down a year | Availability |
| Chain / side by side | Availabilities multiply down / failure chances multiply down | Availability |
| High availability / fault tolerance / DR | Stays up / no interruption / recovers after | High availability |
| RPO / RTO | Data you may lose / time you may be down | Disaster recovery |
| Backup and restore, pilot light, warm standby, multi-site | DR strategies, cheap and slow to instant | Disaster recovery |
| Metrics, logs, traces | What to watch | Monitoring |
| Alarm | A threshold, a duration, an action | Monitoring |
| CloudWatch / Azure Monitor / Cloud Observability | Monitoring services | Monitoring |
| CloudFormation / ARM, Bicep / Terraform | Infrastructure as code | Infrastructure as code |
| Basic, Developer, Business, Enterprise | AWS support plans: you buy response time | Support plans |
| Six pillars | Operational excellence, security, reliability, performance, cost, sustainability | Well-architected design |
Nothing matches. Try a shorter word.