What cloud computing really is
Explain cloud computing in plain language, map it to your on-premises IIS and SQL Server experience, and reason about capacity, cost and agility like a cloud engineer.
Assumes: .NET 10 LTS ยท AWS SDK for .NET v4 ยท AWS CLI v2 ยท Region us-east-1. Verified against official sources on 2026-11-17.
Learning objectives
By the end of this lesson you can:
- Define cloud computing in one sentence and name the five essential characteristics from the NIST definition.
- Explain the six advantages of cloud computing as AWS frames them, using a banking example for each.
- Contrast capital expenditure (CapEx) with operational expenditure (OpEx) and explain why that shift matters to a bank's finance team.
- Distinguish scalability from elasticity, and explain why "stop guessing capacity" changes how you design a month-end batch job.
- Run read-only AWS CLI commands that show the global footprint you now rent, and a small C# model that compares fixed and pay-as-you-go capacity.
Prerequisites
- Module 0: Orientation & a safe learning environment, especially Secure your AWS account and Cost guardrails
- Tools: AWS CLI v2 with the
learn-sandboxSSO profile, .NET 10 SDK (only for the optional code) - No AWS knowledge is assumed
Why this matters in banking
Picture Harbourline Bank's statement run. On the first working day of every month the bank generates PDF statements for about 1.2 million customers. For 28 days a year the batch servers are busy; for the other 337 days they sit at 5% CPU. The infrastructure team bought them to survive the peak, so the bank paid for peak capacity all year โ plus the power, cooling, data-centre floor space, hardware support contracts and the people who patch them.
Then, one January, a regulatory change adds a new fee-disclosure page to every statement. The run grows by 40% and no longer fits in its overnight window. Buying servers takes eight to twelve weeks: business case, procurement, delivery, racking, imaging, security sign-off. The statements go out late, and the complaints team has a bad month.
Cloud computing is, above all, an answer to that story. It does not make the work disappear, and it does not make a bank automatically safer or cheaper. It changes when you commit money, how fast you can get capacity and who does the undifferentiated work underneath your application. Every later module โ networking, security, data pipelines, disaster recovery โ builds on the mental model in this lesson.
The idea in plain language
AWS defines cloud computing as on-demand delivery of IT resources over the internet with pay-as-you-go pricing. Unpack that sentence:
- On demand โ you ask for a server, a database or a storage bucket through a web page or an API call, and you get it in seconds or minutes. No ticket to the infrastructure team, no purchase order.
- IT resources โ not only virtual machines. Storage, databases, queues, machine-learning models, identity services, even entire managed applications.
- Over the internet โ you reach the provider's data centres over the network. (For banks this is often a private network link, which lesson 3 covers.)
- Pay as you go โ you pay for what you use, measured per second, per hour, per request or per gigabyte, and you stop paying when you delete it.
Analogy: Running your own data centre is like owning a fleet of delivery vans. You buy enough vans for the Christmas peak, insure and service them all year, and most of them sit in the depot in February. Cloud computing is like a van-hire network with depots in every city: you hire 3 vans in February and 40 in December, pay by the day, and the hire company does the servicing. You still choose the routes, pack the parcels and lock the doors โ that part stays your job.
The NIST definition: five essential characteristics
The most widely quoted vendor-neutral definition comes from the US National Institute of Standards and Technology, NIST SP 800-145 (September 2011). It says a service is "cloud" when it has five characteristics. They are useful as a checklist when someone calls a hosted virtual machine farm "our private cloud".
| Characteristic | Plain meaning | On-premises reality at a typical bank | AWS example |
|---|---|---|---|
| On-demand self-service | You provision resources yourself, without a human at the provider | A ServiceNow ticket, then a wait of days or weeks | aws s3api create-bucket returns in about a second |
| Broad network access | Resources are reachable over standard networks from many kinds of clients | Reachable only from inside the corporate network | Console, CLI, SDK, all over HTTPS |
| Resource pooling | The provider serves many customers from shared, pooled hardware; you don't know (or care) which physical host you are on | Each application team owns "its" servers | Many AWS customers share the same physical fleet, isolated by the hypervisor and service controls |
| Rapid elasticity | Capacity can grow and shrink quickly, often automatically | Capacity is fixed until the next hardware refresh | EC2 Auto Scaling adds instances when queue depth rises |
| Measured service | Usage is metered, visible and billed | Costs are allocated by spreadsheet once a year | Cost Explorer shows spend per service, per day, per tag |
NIST also defines three service models (IaaS, PaaS, SaaS โ next lesson) and four deployment models (private, community, public, hybrid โ lesson 3).
The six advantages of cloud computing (AWS's framing)
The AWS Overview of Amazon Web Services whitepaper groups the benefits into six advantages. You will meet them on the CLF-C02 exam, so learn the names โ but more importantly, learn the honest banking interpretation of each.
| Advantage | What it means | Harbourline Bank example | The honest caveat |
|---|---|---|---|
| Trade fixed expense for variable expense | Pay only when you consume, instead of buying data centres and servers up front | The statement run costs money for a few hours a month, not all year | Variable costs need active management; idle resources still bill |
| Benefit from massive economies of scale | The provider aggregates demand from a very large number of customers and passes on lower unit costs | AWS buys hardware, power and network at a scale no single bank can | Lower unit prices do not guarantee a lower bill if you over-provision |
| Stop guessing capacity | Scale up and down within minutes instead of sizing for a guessed peak | The January fee-disclosure change needs 40% more compute for one night: add it, then remove it | Account-level service quotas still exist and sometimes need raising in advance |
| Increase speed and agility | New resources are a click or an API call away, so experiments are cheap and fast | A developer can test a new fraud rule on a copy of last month's (synthetic) data the same afternoon | Speed without guardrails creates sprawl; you need accounts, budgets and IaC (Module 0, Module 10) |
| Stop spending money running and maintaining data centres | Focus on differentiating work, not racking and powering servers | The platform team stops replacing failed disks and starts building a payments platform | You still own architecture, security configuration, patching of anything you run on VMs, and operations |
| Go global in minutes | Deploy into Regions around the world with little effort | Serve the EU subsidiary from Frankfurt or Dublin while UK data stays in London | Data-residency, outsourcing and privacy rules limit where a bank may go; "can" is not "may" |
CapEx versus OpEx
- Capital expenditure (CapEx) is money spent up front on assets the company owns and depreciates over years: servers, storage arrays, data-centre fit-out.
- Operational expenditure (OpEx) is ongoing spending on services consumed: electricity, support contracts, and โ in the cloud โ the monthly AWS bill.
Moving to the cloud shifts much of IT spend from CapEx to OpEx. A bank's finance team cares because it changes budgeting (a monthly bill that moves with usage), depreciation schedules, and the business case for new products (a pilot can start small and stop cleanly). It also changes accountability: in a CapEx world a mistake costs a one-off overspend; in an OpEx world a forgotten resource keeps costing money every hour until someone notices. That is why Cost guardrails came before any hands-on work.
AWS does let you trade some flexibility back for a lower price โ Savings Plans and Reserved Instances are commitments that look a little like CapEx. Pricing models covers when that makes sense.
Scalability versus elasticity
These two words are often used interchangeably. They are not the same, and the exams test the difference.
- Scalability is the ability of a system to handle more load by adding resources โ either bigger machines (vertical scaling, "scale up") or more machines (horizontal scaling, "scale out").
- Elasticity is scaling automatically and in both directions, closely tracking demand, so you are not paying for idle capacity afterwards.
A SQL Server box you can upgrade to a larger VM over a weekend is scalable. A fleet of statement-rendering workers that grows from 2 to 40 when a queue fills up and shrinks back to 2 when it drains is elastic. Elasticity is where the money is saved โ and it requires the application to be designed for it (stateless workers, work held in a queue, no local disk state), which Cloud architecture fundamentals explains.
What problem the AWS service solves
This lesson is about cloud computing in general rather than one service, so think of AWS as the product. It exists to solve three problems that every bank's infrastructure team knows:
- Lead time. Physical capacity takes weeks to months to arrive. AWS turns that into minutes.
- Utilisation. On-premises fleets are sized for the peak and sit mostly idle. AWS lets you pay for the curve, not the peak.
- Undifferentiated heavy lifting. Power, cooling, hardware failure, firmware, hypervisors, physical security and, for managed services, operating systems and database engines. None of that differentiates a bank from its competitors.
Limitations and sharp edges you should know from day one:
- You are still responsible for a lot. Identity, data classification, encryption choices, network exposure and application security stay with you. This is the Shared Responsibility Model.
- Costs are easy to start and easy to forget. Some resources (NAT gateways, load balancers, databases) bill every hour whether or not anyone uses them.
- Quotas exist. Each account has default service quotas (for example, how many vCPUs of On-Demand instances you can run). They protect you and AWS from runaway usage, and they sometimes need raising before a big event.
- Not every service is in every Region, and a few services close to new customers over time (for example, AWS App Runner closed to new customers in 2026). Always check availability before designing around a service.
- Regulation does not move to the cloud by itself. Outsourcing rules, data-protection law and operational-resilience requirements still apply to the bank, not to AWS.
Microsoft / Azure comparison
| You know (Microsoft/Azure) | AWS equivalent | What's different |
|---|---|---|
| Your own data centre with Hyper-V or VMware hosts | AWS Regions and Availability Zones running the AWS Nitro-based fleet | You never see or touch the host; you rent capacity through APIs |
| Ordering a new physical server for IIS | Launching an EC2 instance, or better, a managed service such as Fargate or Lambda | Minutes instead of weeks; you can delete it the same day |
| Annual capacity planning spreadsheets | Auto Scaling, serverless services, pay-per-request pricing | You plan limits and budgets, not boxes |
| Azure subscription billing and Cost Management | AWS Billing and Cost Management, Cost Explorer, Budgets | Same pay-as-you-go idea; AWS charges many services per second or per request |
| Azure's own "benefits of cloud" marketing (CapEx to OpEx, scale) | AWS's "six advantages of cloud computing" | Same economics; the exam wants AWS's wording |
| Azure regions and availability zones | AWS Regions and Availability Zones | Very similar concepts, covered in AWS global infrastructure |
If you already moved workloads to Azure, nothing in this lesson is philosophically new. What is new is the vocabulary, the account model and the fact that AWS has far more fine-grained building blocks โ which makes cost and security discipline even more important.
Architecture
The diagram contrasts the statement run on premises with an elastic cloud design. You don't need to know these services yet; focus on the shape.
Security-boundary view. Moving to the right-hand side does not remove trust boundaries; it changes who enforces which one. AWS secures the physical data centre, the hosts and the managed-service software. Harbourline still decides who can read the statements bucket (IAM), whether it is encrypted with a key the bank controls (KMS), and whether access is logged (CloudTrail). Modules 5 and 14 build these controls properly.
Step-by-step implementation
This is a conceptual lesson, so the "implementation" is a short exploration that makes the abstract ideas concrete. Nothing here creates resources or costs money.
Sign in with your sandbox profile
Open Windows Terminal (PowerShell) and refresh your SSO session from Module 0:
aws sso login --profile learn-sandbox
aws sts get-caller-identity --profile learn-sandboxYou should see your sandbox account ID and an ARN containing AWSReservedSSO_SandboxAdmin. That ARN is a temporary identity: on-demand self-service and measured service in one line.
See the global footprint you now rent
Run the Region commands in the AWS CLI section below. Count how many Regions are listed and notice that some are "not opted in". You did nothing to build any of them โ that is economies of scale and "go global in minutes" made visible.
Model fixed versus elastic capacity
Run the small C# program in Complete code. It compares buying servers for the peak with paying only for the hours you use. Change the peak size and the number of peak hours and watch the break-even point move.
Write your own definitions
In your learning journal, write one sentence each for cloud computing, elasticity and CapEx versus OpEx, in words you would use with a non-technical colleague. If you can't, re-read the plain-language section.
AWS Console instructions
Nothing needs creating, but it is worth a five-minute tour so the ideas feel real:
- Open the AWS access portal URL from Module 0, choose the
cea-sandboxaccount and SandboxAdmin โ Management console. - Look at the Region selector at the top right. Open it and scroll: each entry is a separate set of data centres you can deploy to.
- Search for Billing and Cost Management in the search bar. In the sandbox account you may see little or nothing; billing is consolidated in the management account. That separation is deliberate and is covered in AWS accounts and Organizations.
- Search for Service Quotas โ AWS services โ Amazon Elastic Compute Cloud (Amazon EC2). Find "Running On-Demand Standard (A, C, D, H, I, M, R, T, Z) instances". This is the default ceiling on how many vCPUs you can run โ elasticity, but with a safety rail.
AWS CLI
All commands below are read-only and free. They use the learn-sandbox profile and Windows PowerShell quoting.
# List the Regions available to this account, including whether you are opted in.
# ec2:DescribeRegions is a read-only call; it is exempt from the Module 0 Region guardrail.
aws ec2 describe-regions --all-regions --query "Regions[].[RegionName,OptInStatus]" --output table --profile learn-sandbox
# The Account Management API gives the same picture with clearer status values (ENABLED, DISABLED, ENABLED_BY_DEFAULT).
aws account list-regions --query "Regions[].[RegionName,RegionOptStatus]" --output table --profile learn-sandbox
# AWS publishes its global infrastructure as public Systems Manager parameters.
# Count the services AWS lists (this path is queryable from us-east-1 and other major Regions).
aws ssm get-parameters-by-path --path /aws/service/global-infrastructure/services --region us-east-1 --query "length(Parameters)" --profile learn-sandbox
# Which Regions offer AWS Lambda? (Replace lambda with another service code to compare.)
aws ssm get-parameters-by-path --path /aws/service/global-infrastructure/services/lambda/regions --region us-east-1 --query "Parameters[].Value | sort(@)" --profile learn-sandboxWhat to notice:
describe-regions --all-regionsshows Regions you have not enabled. Newer Regions are opt-in; older ones are enabled by default. Opting in is an account-level decision a bank controls centrally.- The
ssmcommands read public parameters owned by AWS.get-parameters-by-pathreturns at most 10 results per page, but the AWS CLI follows the pagination for you, so the count is the full list. - If the
ssmcommand fails in your sandbox with an access-denied error mentioning a service control policy, check that you passed--region us-east-1; the guardrail only allows us-east-1 and eu-west-2.
Infrastructure as Code
Infrastructure as Code (IaC) is itself a cloud benefit: because everything is an API, a whole environment can be described in a text file and recreated identically. This tiny Terraform configuration creates nothing; it only reads the Regions and Availability Zones, which is a safe first taste of the tool you will use from Module 2 onwards.
terraform {
required_providers {
aws = {
source = "hashicorp/aws"
version = "~> 6.0"
}
}
}
provider "aws" {
region = "us-east-1"
profile = "learn-sandbox"
}
# Every Region enabled for this account.
data "aws_regions" "enabled" {}
# Availability Zones in the provider's Region (us-east-1).
data "aws_availability_zones" "here" {
state = "available"
}
output "enabled_region_count" {
value = length(data.aws_regions.enabled.names)
}
output "us_east_1_zones" {
value = data.aws_availability_zones.here.names
}terraform init
terraform plan # shows the outputs it would produce; nothing is createdComplete code
A small, self-contained C# program that models the statement run. It does not call AWS and uses illustrative unit prices (not real AWS prices) so you can focus on the shape of the economics. Real prices come later, in Pricing models and the Pricing Calculator lab.
<Project Sdk="Microsoft.NET.Sdk">
<PropertyGroup>
<OutputType>Exe</OutputType>
<TargetFramework>net10.0</TargetFramework>
<Nullable>enable</Nullable>
<ImplicitUsings>enable</ImplicitUsings>
<TreatWarningsAsErrors>true</TreatWarningsAsErrors>
<RootNamespace>Harbourline.Learning.CapacityModel</RootNamespace>
</PropertyGroup>
</Project>// Compares two ways to provide capacity for a monthly statement run.
// All prices are ILLUSTRATIVE ONLY - they are not AWS prices.
using System.Globalization;
var culture = CultureInfo.GetCultureInfo("en-GB");
var workload = new Workload(
BaselineServers: 2, // servers needed every hour of the year
PeakServers: 40, // servers needed during the statement run
PeakHoursPerMonth: 6); // how long the run lasts each month
var owned = new OwnedCapacity(
ServerPurchasePrice: 6_000m, // GBP per server, paid up front
LifetimeYears: 4, // depreciation period
YearlyRunCostPerServer: 1_200m); // power, cooling, space, support
var rented = new RentedCapacity(PricePerServerHour: 0.20m); // GBP, pay as you go
decimal ownedYearly = owned.YearlyCost(workload.PeakServers);
decimal rentedYearly = rented.YearlyCost(workload);
Console.WriteLine($"Owned (sized for peak of {workload.PeakServers}): {ownedYearly.ToString("C0", culture)} per year");
Console.WriteLine($"Rented (baseline {workload.BaselineServers}, peak {workload.PeakHoursPerMonth} h/month): {rentedYearly.ToString("C0", culture)} per year");
Console.WriteLine($"Average utilisation of owned fleet: {workload.AverageUtilisation(workload.PeakServers):P1}");
public sealed record Workload(int BaselineServers, int PeakServers, int PeakHoursPerMonth)
{
public const int HoursPerYear = 8_760;
public decimal ServerHoursPerYear()
{
decimal peakHoursPerYear = PeakHoursPerMonth * 12m;
decimal baselineHours = BaselineServers * (HoursPerYear - peakHoursPerYear);
decimal peakHours = PeakServers * peakHoursPerYear;
return baselineHours + peakHours;
}
public decimal AverageUtilisation(int ownedServers) =>
ServerHoursPerYear() / (ownedServers * (decimal)HoursPerYear);
}
public sealed record OwnedCapacity(decimal ServerPurchasePrice, int LifetimeYears, decimal YearlyRunCostPerServer)
{
public decimal YearlyCost(int servers) =>
servers * (ServerPurchasePrice / LifetimeYears + YearlyRunCostPerServer);
}
public sealed record RentedCapacity(decimal PricePerServerHour)
{
public decimal YearlyCost(Workload workload) =>
decimal.Round(workload.ServerHoursPerYear() * PricePerServerHour, 2);
}Run it with dotnet run. With the sample numbers the output is:
Owned (sized for peak of 40): ยฃ108,000 per year
Rented (baseline 2, peak 6 h/month): ยฃ4,051 per year
Average utilisation of owned fleet: 5.8%Code explained line by line
using System.Globalization;andCultureInfo.GetCultureInfo("en-GB")โ format money as pounds. The project usesdecimaleverywhere because money must never bedouble(see the course conventions).new Workload(BaselineServers: 2, PeakServers: 40, PeakHoursPerMonth: 6)โ the shape of the demand curve: a small steady base and a short, tall peak. Named arguments keep the intent readable.OwnedCapacityโ up-front purchase price spread over a depreciation lifetime (the CapEx part) plus a yearly running cost per server (the hidden OpEx of owning hardware).RentedCapacity(PricePerServerHour: 0.20m)โ a single pay-as-you-go hourly rate. Themsuffix makes it adecimalliteral.owned.YearlyCost(workload.PeakServers)โ when you own hardware you must buy for the peak, so the cost depends on 40 servers, not on how many hours they work.rented.YearlyCost(workload)โ when you rent, cost depends on server-hours actually used.ServerHoursPerYear()โ baseline servers for the non-peak hours plus peak servers for the peak hours. This is the area under the demand curve.AverageUtilisationโ used hours divided by owned capacity-hours. A low number is money spent on idle hardware.decimal.Round(..., 2)โ round to pence only at the end, never in the middle of a calculation.
What the model leaves out (on purpose): data transfer, storage, licences, staff time, the cost of building elastic software, and commitment discounts. Real comparisons need all of those; the AWS Pricing Calculator and your finance team's model are the right tools. The point of the toy is the shape: the spikier the demand curve, the bigger the gap between paying for the peak and paying for the curve.
Security considerations
- The cloud is not "someone else's problem". AWS secures the infrastructure; you secure your identities, data and configuration. Read the Shared Responsibility Model lesson before you design anything real.
- Agility needs guardrails. The same speed that lets a developer create a server in a minute lets them create a public S3 bucket in a minute. Banks pair cloud agility with account boundaries, service control policies, mandatory encryption and automated detection (Modules 5 and 15).
- No long-lived keys. Everything in this course uses short-lived IAM Identity Center sessions. The CLI commands above only work while your SSO session is valid โ that is a feature.
- Data classification first. Before any real customer data goes to the cloud, a bank classifies it, decides permitted Regions and agrees controls with risk and data-protection teams.
โ Learning shortcut โ not for production
The capacity model above compares a single price per server-hour against a single purchase price. Real business cases include licensing, data transfer, staff, migration effort, exit costs and commitment discounts, and in a bank they are reviewed by finance, architecture and procurement. Use the toy to understand the shape of cloud economics, never to justify a migration.
โ Learning shortcut โ not for production
You are using a sandbox with AdministratorAccess to run read-only commands. In a bank, a role that only needs to read Region and quota information would get a narrow read-only policy, and nobody would explore with an administrator session.
Banking use case
Harbourline Bank's architecture board approves a cloud "statement platform" pilot with three conditions:
- Elastic by design. Statement jobs go onto a queue, workers are stateless containers, and the worker count follows the queue length. The bank stops paying for 38 idle servers for 337 days of the year.
- Data stays in approved locations. Statements for UK customers are produced and stored in the London Region (eu-west-2). The EU subsidiary uses an EU Region. The data-protection officer signs off the design; nobody assumes "AWS is compliant, so we are".
- Cost is visible. Every resource carries
Project,EnvironmentandOwnertags, and a budget alerts the product owner at 80% of forecast spend.
The pilot does not migrate the core banking system. The statements are rendered from data exported nightly from the on-premises core โ a hybrid pattern explored in lesson 3. Whether this arrangement counts as material outsourcing under the bank's regulators is decided by the bank's risk, compliance and legal teams, not by the engineers.
Common mistakes
- "Cloud is always cheaper" โ people compare an on-premises server's purchase price with an hourly rate โ a lifted-and-shifted, always-on, oversized estate can cost more in the cloud. Savings come from elasticity, right-sizing, managed services and turning things off.
- Treating elasticity as automatic โ "it's in the cloud, so it scales" โ a stateful application with session data on local disk cannot scale out. Elasticity is an architecture property; design for it.
- Confusing scalability and elasticity โ both mean "bigger" in casual speech โ elasticity also means smaller again, automatically.
- Forgetting the meter is running โ on premises an idle server costs nothing extra this month โ in the cloud, an idle load balancer or database bills every hour. Tag, budget and tear down.
- Assuming the provider takes over compliance โ "AWS has certifications" โ AWS attestations cover AWS's part; the bank remains accountable for its own use of the service.
- Calling a virtualised data centre "private cloud" without self-service or metering โ it fails the NIST characteristics, and the business still waits weeks for capacity.
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
Error when retrieving token from sso: Token has expired and refresh failed | Your IAM Identity Center session has expired | aws sso login --profile learn-sandbox |
The config profile (learn-sandbox) could not be found | Module 0 tooling setup not finished, or a typo in the profile name | Check %USERPROFILE%\.aws\config; re-run aws configure sso (see CLI with SSO profiles) |
AccessDeniedException ... explicit deny in a service control policy on the ssm command | You queried a Region the sandbox guardrail blocks | Add --region us-east-1 |
aws account list-regions returns AccessDenied | Your role lacks account:ListRegions (it is included in AdministratorAccess) | Use aws ec2 describe-regions --all-regions instead, or check the permission set |
terraform plan says "No valid credential sources found" | Terraform cannot see your SSO session | Run aws sso login --profile learn-sandbox first; make sure the provider block sets profile |
C# program shows ยค instead of ยฃ | The console cannot render the currency symbol | Run [Console]::OutputEncoding = [Text.Encoding]::UTF8 in PowerShell, or use Windows Terminal |
Cost and cleanup
Cost: $0. Every command in this lesson is a read-only API call (DescribeRegions, ListRegions, GetParametersByPath on public parameters, GetCallerIdentity), and these carry no charge. Terraform data sources create nothing. The C# program runs locally.
Cleanup: remove the local Terraform working files if you created them.
# Run inside the folder where you ran terraform init
Remove-Item -Recurse -Force .terraform, .terraform.lock.hcl -ErrorAction SilentlyContinueOptionally confirm nothing tagged for the course exists yet (it should return an empty list at this point in the course):
aws resourcegroupstaggingapi get-resources --tag-filters Key=Project,Values=cea-lab --query "ResourceTagMappingList[].ResourceARN" --profile learn-sandboxExercises
- ๐ข Guided: Run the four CLI commands. Record (a) how many Regions exist, (b) how many are enabled for your account, (c) how many services AWS lists, and (d) how many Regions offer Lambda. Then run the C# model with the default numbers.
- ๐ก Challenge: Change the C# model so that the peak lasts 72 hours a month and the baseline is 20 servers. At what point does owning start to look competitive? Explain in two sentences what kind of workload that represents and what AWS pricing option (from Pricing models) you would investigate for it.
- ๐ด Live problem: Your head of infrastructure forwards an email from the CFO: "The vendor says moving to AWS will cut our infrastructure cost by 40%. Can you confirm by Friday?" Write the reply. You have no detailed inventory, no utilisation data and no pricing yet.
Assignment
Create docs/m01-cloud-business-case.md in your learning-journal repository: a one-page note for a non-technical manager that explains the six advantages of cloud computing using one Harbourline Bank example each, plus a "what doesn't change" section (at least four responsibilities that stay with the bank). Include the output of your C# model with your own numbers and a sentence on why the numbers are illustrative.
Solutions
The exact numbers change as AWS adds Regions and services, so there is no single right answer. You should see a few dozen commercial Regions in describe-regions --all-regions, with several marked not-opted-in; a smaller number enabled in list-regions; a few hundred service codes from the ssm path; and Lambda in nearly every Region. Why it matters: you just used on-demand self-service and broad network access against a global footprint you did not build โ the definition of cloud computing in practice.
With BaselineServers: 20, PeakServers: 40 and PeakHoursPerMonth: 72, the model uses 192,480 server-hours a year (most of them from the 20 always-on servers), rented capacity costs about ยฃ38,496 a year against ยฃ108,000 for owned, and owned-fleet utilisation rises from 5.8% to about 54.9%. The gap narrows sharply; push the baseline all the way to 40 servers and rented reaches about ยฃ70,080 โ owned only wins in this toy if you also raise the hourly price or lower the hardware cost, which is exactly why real decisions need real prices. This describes a steady, predictable workload โ the kind for which Savings Plans or Reserved Instances (commitment discounts) are designed, while the spiky part still benefits from On-Demand or Spot. Why: cloud economics are strongest for variable demand; for flat demand the advantage comes from commitment pricing, managed services and avoided data-centre costs rather than from elasticity alone.
A good reply neither confirms nor dismisses the 40%. It explains that savings depend on how workloads move (rehost versus re-architect), on utilisation today (spiky workloads save most), on licensing (for example SQL Server and Windows), on data transfer and on operating model changes. It proposes a short discovery: an inventory of servers with CPU and memory utilisation, a classification of workloads by migration strategy, a priced estimate in the AWS Pricing Calculator for a representative slice, and a review with finance. It names the non-financial benefits (lead time, resilience options) and the new costs (training, landing zone, dual running during migration). Why: in a bank, an unqualified number in an email becomes a commitment; your job is to replace it with a method.
Knowledge check
Answer: On-demand delivery of IT resources (compute, storage, databases and more) over a network, with pay-as-you-go pricing. A good answer mentions both self-service on demand and metered payment; "someone else's computer" misses both.
Answer: On-demand self-service (and usually rapid elasticity and measured service too). Resource pooling through virtualisation alone does not make it a cloud.
Answer: Scalability is the ability to handle more load by adding resources. Elasticity is adding and removing resources automatically as demand changes, so you don't pay for idle capacity afterwards.
Answer: Trade fixed expense for variable expense (the CapEx to OpEx shift). "Economies of scale" is about the provider's lower unit costs, not about when you pay.
Answer: Any two of: identity and access management, data classification and encryption choices, network exposure configuration, application security, patching guest operating systems on EC2, regulatory accountability (outsourcing, data protection, operational resilience), cost management.
Interview questions
- "Explain cloud computing to our Head of Operations in 30 seconds." โ Instead of buying and running servers for our peak, we rent exactly the capacity we need, when we need it, from a provider that runs the data centres. We pay monthly for what we use, we get capacity in minutes instead of months, and we still own the security and correct use of our data.
- "Is the cloud cheaper?" โ It can be, but not automatically. Savings come from elasticity, right-sizing, managed services and switching things off; a straight copy of an always-on estate can cost more. I would measure utilisation first and price a representative workload.
- "What's the difference between scalability and elasticity, and why does it matter for a month-end batch?" โ Scalability is the ability to grow; elasticity is growing and shrinking automatically with demand. For a month-end batch, elasticity means 40 workers for six hours and two the rest of the month, so you pay for the curve, not the peak โ but only if the batch is designed as stateless workers pulling from a queue.
- "What does CapEx to OpEx mean for a bank's IT budget?" โ Spend moves from depreciating assets bought up front to a variable monthly bill. Budgeting becomes continuous, small pilots become cheap to start and stop, and cost control becomes an engineering responsibility through tagging, budgets and teardown.
Certification practice
A retail bank currently buys servers to handle its busiest week of the year. For the rest of the year most of these servers are idle. Which advantage of cloud computing most directly addresses this problem?
A. Benefit from massive economies of scale B. Stop guessing capacity C. Go global in minutes D. Stop spending money running and maintaining data centres
Answer: B. The issue is sizing for a guessed peak; the cloud lets the bank add capacity for the busy week and remove it afterwards. A is about the provider's lower unit prices, not about matching capacity to demand. C is about geographic reach. D is about avoiding data-centre operations work, which is related but does not address idle capacity directly.
A finance director asks how moving workloads to AWS changes the company's IT spending. Which statement is most accurate?
A. Spending moves from operational expenditure to capital expenditure B. Spending moves largely from up-front capital expenditure to variable operational expenditure C. All IT spending is eliminated because AWS owns the hardware D. Spending becomes fixed for the life of the contract
Answer: B. Cloud replaces up-front hardware purchases with pay-as-you-go charges. A reverses the direction. C is wrong because you still pay for the resources you consume. D is wrong because On-Demand pricing varies with usage; only optional commitments such as Savings Plans fix part of the spend.
A payments team runs a fleet of workers that automatically grows from 4 to 60 instances when a queue backs up and shrinks to 4 when the queue empties. Which cloud concept does this demonstrate?
A. High availability B. Elasticity C. Fault tolerance D. Vertical scaling
Answer: B. Automatically adding and removing capacity to match demand is elasticity. A and C concern staying available when components fail, which this scenario does not describe. D means making a single machine bigger, whereas the fleet adds more machines (horizontal scaling).
A bank runs a nightly reconciliation job that needs 32 vCPUs for 90 minutes and almost nothing for the rest of the day. The job currently runs on two large always-on EC2 instances. The bank wants to reduce cost without changing the job's runtime. Which approach is most cost-effective?
A. Purchase a three-year Reserved Instance for the two existing instances B. Move the job to compute that is provisioned only while the job runs, triggered on a schedule C. Increase the instance size so the job finishes faster and leave the instances running D. Move the instances to a different Availability Zone
Answer: B. Provisioning compute only for the 90-minute window means paying for about 6% of the day instead of 100%. A discounts capacity that is idle 94% of the time โ you would still pay for idle hours. C increases cost and still pays for idle time. D changes location, not cost.
Completion checklist
- I can define cloud computing in one sentence and name the five NIST characteristics
- I can name the six advantages of cloud computing and give a banking example of each
- I can explain CapEx versus OpEx and scalability versus elasticity to a colleague
- I ran the read-only CLI commands and the C# capacity model
- I confirmed nothing tagged
Project=cea-labexists yet, and removed local Terraform files
Key takeaways
- Cloud computing is on-demand, metered IT resources over a network; NIST's five characteristics are a good test of whether something really is "cloud".
- AWS frames the benefits as six advantages: variable expense, economies of scale, stop guessing capacity, speed and agility, no data-centre upkeep, and global reach.
- The biggest savings come from elasticity โ paying for the demand curve rather than the peak โ and elasticity has to be designed into the application.
- The meter is always running: cost discipline and teardown are engineering responsibilities.
- Moving to the cloud moves work, not accountability; security configuration, data protection and regulatory obligations stay with the bank.
Further reading
- What is cloud computing? (AWS)
- Six advantages of cloud computing โ Overview of Amazon Web Services whitepaper
- Types of cloud computing โ Overview of Amazon Web Services whitepaper
- NIST SP 800-145: The NIST Definition of Cloud Computing
- Public parameters for AWS global infrastructure (Systems Manager Parameter Store)
- AWS Certified Cloud Practitioner (CLF-C02) exam guide
- Course pages: CLF-C02 study guide ยท Glossary
Independent learning resource โ not affiliated with, endorsed or sponsored by Amazon Web Services. AWS and related marks are trademarks of Amazon.com, Inc. or its affiliates. All banking data in examples is fictional. Sample architectures are for learning and are not automatically compliant with GDPR, PCI DSS or any banking regulation.
Module 1: Cloud Computing & AWS Fundamentals
Learn what the cloud really is, how AWS is organised (Regions, Availability Zones, accounts) and how you talk to it from the console, the CLI and C#. You finish by calling AWS from a .NET console app using short-lived SSO credentials, deploying something small and tearing it down.
IaaS, PaaS and SaaS
Tell the cloud service models apart, place AWS and Azure services on the spectrum from virtual machines to serverless, and choose a model for a banking workload by what you are willing to operate.