Lab: price a banking web app three ways in the AWS Pricing Calculator
Estimate Harbourline Bank's statement portal in the AWS Pricing Calculator with On-Demand, Savings Plans and Spot workers, compare PostgreSQL with SQL Server licensing, and check the totals in C#.
Assumes: .NET 10 LTS (file-based app) ยท AWS Pricing Calculator (public site) ยท Region us-east-1 ยท prices from AWS Price List files published 2026-11-13 to 2026-11-17. Verified against official sources on 2026-11-17.
Learning objectives
By the end of this lab you can:
- Build a grouped estimate in the public AWS Pricing Calculator for a small banking web application.
- Price the same compute three ways โ On-Demand, Compute Savings Plan and Spot โ and explain why the answer differs for a 24ร7 web tier and a 3-hour nightly batch.
- Show finance colleagues how much a SQL Server License Included database adds compared with PostgreSQL.
- Share and export an estimate safely, without leaking anything confidential.
- Recompute the estimate in C# with
decimalarithmetic to check the calculator and run "what if" scenarios.
Prerequisites
- Pricing models โ On-Demand, Savings Plans, RIs, Spot, licensing
- Free Tier and billing โ for the always-free allowances used below
- A web browser. No AWS account is needed for the public calculator.
- Optional: .NET 10 SDK, to run the checking program
Why this matters in banking
Before Harbourline Bank's architecture review board approves anything, it asks for a cost estimate with options. "It will cost about USD 500 a month" is not enough: the board wants to know the cost of the cheapest safe option, what a commitment would save and what it would risk, and how much the database licence choice matters. Banks also have annual budgeting cycles; an engineer who can produce a clear, shareable, defensible estimate in an hour saves weeks of back-and-forth with finance.
This lab gives you that skill on a realistic, fictional workload.
The idea in plain language
The AWS Pricing Calculator is a free website where you describe what you plan to run (instance types, hours, storage, requests) and it shows the monthly and annual cost. You can group services, compare pricing options, save a link and export a CSV or PDF.
There are two calculators:
Public Pricing Calculator (calculator.aws) | In-console Pricing Calculator (Billing and Cost Management) | |
|---|---|---|
| Needs an AWS account | No | Yes |
| Uses your discounts and existing commitments | No โ public list prices | Yes (workload estimates can use before-discount, after-discount and after-commitment rates) |
| Can model your whole consolidated bill | No | Yes โ bill estimates, management or standalone accounts only |
| Price | Free | Workload estimates free; bill estimates: 5 free per month, then USD 2 each |
This lab uses the public calculator, so it is free and needs no account.
Analogy: The calculator is the quote you get from a builder before work starts. It lists every material and hour of labour, so you can ask "what if we use cheaper tiles?" before anyone picks up a hammer. It is still a quote โ the final invoice depends on what you actually use.
What problem the AWS service solves
- Model before you build: compare architectures and pricing options with no resources running.
- Transparent maths: each service shows its calculation, so reviewers can check every assumption.
- Shareable: a saved estimate becomes a public link; exports go into business cases.
Limitations to be honest about:
- It only prices what you enter. Forgotten items (NAT Gateway, CloudWatch logs, public IPv4 addresses, backups, support plan, data transfer between AZs) are simply missing.
- It uses list prices from the AWS Price List API. If a service pricing page and the calculator disagree, AWS charges the pricing-page price.
- Saved estimates are stored on AWS public servers and shared links work for anyone who has them. Links created since 31 May 2023 stay valid for one year.
- Spot estimates are based on historical averages; real Spot prices move every hour.
- Free Tier allowances are not always applied automatically, so check each service's free-tier option.
Microsoft / Azure comparison
| You know (Microsoft/Azure) | AWS equivalent | What's different |
|---|---|---|
| Azure Pricing Calculator | AWS Pricing Calculator (calculator.aws) | Very similar workflow: add services, group, export. AWS share links expire after a year |
| Azure Cost Management "cost analysis" forecast | In-console Pricing Calculator bill estimates, Cost Explorer forecasts | AWS can layer modelled changes onto last month's real consolidated bill |
| Azure Hybrid Benefit toggle in the calculator | Licence model choice (License Included, BYOL on Dedicated Hosts, RDS Bring Your Own Media) | No single "hybrid benefit" switch: licence rules differ for Windows Server and SQL Server |
| Azure savings plan / reservation options per VM | EC2 pricing strategy: On-Demand, Savings Plans, Spot, RIs | Savings Plans are bought as USD per hour, not per VM |
| Excel business case | Exported CSV/PDF estimate | Keep the share link in the business case so reviewers can open the details |
Architecture
The fictional Harbourline statement portal: customers log in, view and download monthly PDF statements. PDFs are rendered overnight by a worker fleet and stored in S3.
Assumptions used throughout (write them into your estimate description so reviewers see them):
| Component | Assumption |
|---|---|
| Region | US East (N. Virginia) us-east-1 โ the course default. A UK bank would price eu-west-2 as well |
| Web tier | 2 ร m7i.large, Linux, 730 hours/month each |
| Load balancer | 1 Application Load Balancer, average 1 LCU |
| Database | RDS db.m6i.large, Multi-AZ, 100 GB gp3 |
| Statements | 500 GB in S3 Standard; 100,000 PUTs and 1,000,000 GETs per month |
| CDN | CloudFront, 200 GB to the internet, 2 million HTTPS requests per month |
| Batch | 10 ร m7i.large, 3 hours per night, 30 nights = 900 instance-hours/month |
| Excluded | Data transfer between AZs, NAT Gateway, CloudWatch, AWS WAF, AWS Backup, KMS keys, support plan, taxes |
Step-by-step implementation
Button names below follow the AWS Pricing Calculator User Guide (checked 2026-11-17). The calculator UI changes from time to time; if a label differs, look for the closest equivalent.
Create the estimate and groups
- Open calculator.aws and choose Create estimate.
- You land on Add service. Before adding anything, go to My estimate and choose Create group three times:
Web tier,Data,Batch workers. - Name the estimate (edit the title at the top of My estimate)
Harbourline statement portal - fictional - us-east-1.
Groups let you see subtotals by architecture layer โ exactly how a review board reads a cost.
Price the web tier On-Demand
- Select the
Web tiergroup in the left panel, choose Add service, search Amazon EC2, then Configure. - Description:
Web tier 2x m7i.large On-Demand. Region: US East (N. Virginia). - Operating system Linux; workload: constant usage; Number of instances: 2; instance type m7i.large.
- Pricing strategy / payment option: On-Demand โ VERIFY the exact label in the current UI.
- EBS: 30 GB gp3 per instance (or 0 to keep the comparison pure โ note which you chose).
- Choose Save and add service.
Record the monthly figure in your results table (below).
Add the same web tier with a Compute Savings Plan
Add Amazon EC2 again in the same group with description Web tier 2x m7i.large Compute SP 1yr No Upfront. Choose Compute Savings Plans, 1 year, No upfront. Save.
You now have two lines for the same servers. You will delete one before sharing, but keeping both side by side makes the comparison obvious.
Add the load balancer
In Web tier, add Elastic Load Balancing โ Application Load Balancer: 1 load balancer, and enter traffic so that the average comes out at about 1 LCU (for example, a small number of new connections per second and a few GB processed per hour). Description: ALB avg 1 LCU.
Price the database twice: PostgreSQL vs SQL Server
In Data, add Amazon RDS for PostgreSQL: db.m6i.large, Multi-AZ, On-Demand, 100 GB gp3. Description RDS PostgreSQL Multi-AZ.
Then add Amazon RDS for SQL Server: db.m6i.large, Multi-AZ, Standard Edition, License Included, On-Demand, 100 GB gp3. Description RDS SQL Server SE LI Multi-AZ (comparison only).
Compare the two lines. This is the most important number in the lab.
Add S3 and CloudFront
In Data, add Amazon S3 โ S3 Standard: 500 GB storage, 100,000 PUT/COPY/POST/LIST requests, 1,000,000 GET requests. Description Statement PDFs.
In Web tier, add Amazon CloudFront: 200 GB data transfer out to the US/Europe and 2,000,000 HTTPS requests. If the calculator offers an always-free tier option, turn it on and note it. CloudFront's pay-as-you-go always-free allowance (1 TB out and 10 million requests per month) covers this whole scenario. CloudFront also offers flat-rate plans, including a USD 0 Free plan with lower allowances; the calculator may model only pay-as-you-go โ VERIFY.
Price the nightly workers three ways
In Batch workers, add Amazon EC2 three times, each with 10 ร m7i.large Linux and usage of 3 hours per day (or 90 hours per month per instance โ the usage field accepts different time units โ VERIFY label):
Batch On-Demandโ On-Demand.Batch Compute SPโ Compute Savings Plan, 1-year No Upfront. Look carefully at the result.Batch Spotโ Spot. The calculator shows a historical-average Spot discount.
Review, share and export
- Open My estimate โ View summary. Expand each group to see subtotals and the Show calculations detail for one EC2 line.
- Delete the comparison lines you don't want in the "recommended" view, or keep them and label them clearly.
- Choose Share โ read the Public server acknowledgment โ Agree and Continue โ Copy public link. Store the link in your journal.
- Choose Export โ CSV (and PDF if you like). The PDF contains the share link.
Check the totals in C#
Run the program in Complete code with the unit prices you saw. Differences of a few cents are rounding; larger differences mean an input differs (for example EBS volumes or LCUs).
Results template (fill this in)
| Line | Group | Pricing option | Your USD/month |
|---|---|---|---|
| Web 2 ร m7i.large | Web tier | On-Demand | |
| Web 2 ร m7i.large | Web tier | Compute SP 1-yr No Upfront | |
| ALB (1 LCU) | Web tier | On-Demand | |
| CloudFront 200 GB / 2M req | Web tier | Pay-as-you-go (free tier) | |
| RDS PostgreSQL Multi-AZ + 100 GB | Data | On-Demand | |
| RDS SQL Server SE LI Multi-AZ + 100 GB | Data | On-Demand | |
| S3 500 GB + requests | Data | On-Demand | |
| Batch 900 instance-hours | Batch | On-Demand | |
| Batch 10 ร m7i.large | Batch | Compute SP 1-yr No Upfront | |
| Batch 900 instance-hours | Batch | Spot |
Sample results (approximate โ for checking only)
These numbers are approximate. They were calculated from the AWS Price List files published between 2026-11-13 and 2026-11-17 (Spot from a single 2026-11-17 snapshot), with no EBS volumes on the EC2 lines. Your calculator figures will differ somewhat, especially for Spot.
| Line | Unit price used | USD/month |
|---|---|---|
| Web 2 ร m7i.large On-Demand | 0.1008/h | 147.17 |
| Web 2 ร m7i.large Compute SP 1-yr NU | 0.07406/h | 108.13 |
| ALB | 0.0225/h + 1 LCU ร 0.008/h | 22.27 |
| CloudFront | Inside always-free allowance | 0.00 |
| RDS PostgreSQL db.m6i.large Multi-AZ + 100 GB | 0.356/h + 0.23/GB-month | 282.88 |
| RDS SQL Server SE LI db.m6i.large Multi-AZ + 100 GB | 1.954/h + 0.23/GB-month | 1,449.42 |
| S3 Standard 500 GB + 100k PUT + 1M GET | 0.023/GB, 0.005 per 1k PUT, 0.0004 per 1k GET | 12.40 |
| Batch On-Demand (900 h) | 0.1008/h | 90.72 |
| Batch Compute SP sized for 10 instances | 0.7406/h committed for all 730 h | 540.64 |
| Batch Spot (900 h) | about 0.0449/h | about 40.41 |
Scenario totals (output of the C# program below):
| Scenario | USD/month | USD/year | vs A |
|---|---|---|---|
| A โ All On-Demand, PostgreSQL | 555.43 | 6,665.20 | โ |
| B โ Compute SP for web, Spot workers, PostgreSQL | 466.08 | 5,592.99 | โ16.1% |
| C โ Mistake: Savings Plan also sized for the 3-hour batch | 966.31 | 11,595.73 | +74.0% |
| D โ All On-Demand, SQL Server Standard License Included | 1,721.97 | 20,663.68 | +210.0% |
What the numbers teach:
- Commit to the baseline only. A Savings Plan is billed every hour. Covering a fleet that runs 3 hours a night means paying for 21 idle hours a day โ scenario C costs far more than plain On-Demand.
- Spot fits the batch. The workers are restartable, so Spot halves their cost.
- The licence dwarfs everything else. The SQL Server Standard licence adds more than USD 1,100 a month โ twice the rest of the application. That is why Module 13 spends time on SQL Server-to-PostgreSQL options and on licensing strategy.
- Database commitments are separate. A Database Savings Plan (1-year, No Upfront, up to 35% off) could reduce the RDS line further once usage is stable.
AWS Console instructions
This lab uses the public website, not the AWS Management Console. If you want to try the in-console version later: Console โ Billing and Cost Management โ Pricing Calculator โ Workload estimates โ Create workload estimate. Stick to workload estimates (free). Bill estimates are limited to 5 free per month and then cost USD 2 each, and they are only available to management or standalone accounts.
AWS CLI
Not needed for this lab: the public calculator has no CLI. To check a single unit price from the terminal, use the Price List commands from Pricing models, for example:
# Confirm the On-Demand price of the web instance used in the estimate (read-only, free).
aws pricing get-products --service-code AmazonEC2 --region us-east-1 --profile learn-sandbox `
--filters Type=TERM_MATCH,Field=instanceType,Value=m7i.large `
Type=TERM_MATCH,Field=regionCode,Value=us-east-1 `
Type=TERM_MATCH,Field=operatingSystem,Value=Linux `
Type=TERM_MATCH,Field=tenancy,Value=Shared `
Type=TERM_MATCH,Field=preInstalledSw,Value=NA `
Type=TERM_MATCH,Field=capacitystatus,Value=Used `
--max-results 1 --query "PriceList[0]" --output textInfrastructure as Code
Not applicable: nothing is deployed in this lab. In later modules you can generate cost estimates from IaC (for example by reviewing a CDK or Terraform plan against these unit prices) before deploying.
Complete code
A .NET 10 file-based app that recomputes the four scenarios with decimal arithmetic. It makes no AWS calls and needs no credentials. Save it as StatementPortalEstimate.cs anywhere and run dotnet run StatementPortalEstimate.cs.
#:property TreatWarningsAsErrors=true
#:property Nullable=enable
// Recomputes the Harbourline statement-portal estimate from unit prices,
// so you can check the AWS Pricing Calculator's totals and try "what if" changes.
// No AWS calls, no credentials, no cost. Replace the unit prices with the ones you saw in the calculator.
using System.Globalization;
const decimal HoursPerMonth = 730m;
// Unit prices, us-east-1, USD (AWS Price List files of 2026-11-13..17; Spot is a 2026-11-17 snapshot).
var p = new UnitPrices(
WebOnDemandHourly: 0.1008m, // m7i.large Linux On-Demand
WebComputeSpHourly: 0.07406m, // m7i.large Compute Savings Plan 1-year No Upfront
WorkerSpotHourly: 0.0449m, // m7i.large Linux Spot (changes constantly)
AlbHourly: 0.0225m,
AlbLcuHourly: 0.008m,
RdsPostgresMultiAzHourly: 0.356m, // db.m6i.large PostgreSQL Multi-AZ
RdsSqlServerSeLiMultiAzHourly: 1.954m, // db.m6i.large SQL Server Standard, License Included, Multi-AZ
RdsMultiAzGp3PerGbMonth: 0.23m,
S3StandardPerGbMonth: 0.023m,
S3PutPer1000: 0.005m,
S3GetPer1000: 0.0004m);
// Workload assumptions for the fictional statement portal.
var w = new Workload(
WebInstances: 2,
AverageLcus: 1m,
DbStorageGb: 100m,
StatementsGb: 500m,
PutRequests: 100_000m,
GetRequests: 1_000_000m,
Workers: 10,
WorkerHoursPerNight: 3m,
NightsPerMonth: 30m);
var shared = Alb(p, w) + S3(p, w); // CloudFront: within the always-free allowance in this scenario (USD 0)
var workerHours = w.Workers * w.WorkerHoursPerNight * w.NightsPerMonth;
var scenarios = new (string Name, decimal Monthly)[]
{
("A All On-Demand, PostgreSQL",
w.WebInstances * p.WebOnDemandHourly * HoursPerMonth + shared + Postgres(p, w) + workerHours * p.WebOnDemandHourly),
("B Compute SP for web, Spot workers, PostgreSQL",
w.WebInstances * p.WebComputeSpHourly * HoursPerMonth + shared + Postgres(p, w) + workerHours * p.WorkerSpotHourly),
("C MISTAKE: SP also sized for the 3-hour batch",
(w.WebInstances + w.Workers) * p.WebComputeSpHourly * HoursPerMonth + shared + Postgres(p, w)),
("D All On-Demand, SQL Server Standard (LI)",
w.WebInstances * p.WebOnDemandHourly * HoursPerMonth + shared + SqlServer(p, w) + workerHours * p.WebOnDemandHourly),
};
var baseline = scenarios[0].Monthly;
var gb = CultureInfo.InvariantCulture;
Console.WriteLine($"{"Scenario",-50} {"USD/month",10} {"USD/year",10} {"vs A",8}");
foreach (var (name, monthly) in scenarios)
{
var delta = (monthly - baseline) / baseline;
Console.WriteLine(string.Create(gb, $"{name,-50} {monthly,10:0.00} {monthly * 12,10:0.00} {delta,8:+0.0%;-0.0%;0.0%}"));
}
static decimal Alb(UnitPrices p, Workload w) => (p.AlbHourly + w.AverageLcus * p.AlbLcuHourly) * HoursPerMonth;
static decimal S3(UnitPrices p, Workload w) =>
w.StatementsGb * p.S3StandardPerGbMonth + w.PutRequests / 1000m * p.S3PutPer1000 + w.GetRequests / 1000m * p.S3GetPer1000;
static decimal Postgres(UnitPrices p, Workload w) => p.RdsPostgresMultiAzHourly * HoursPerMonth + w.DbStorageGb * p.RdsMultiAzGp3PerGbMonth;
static decimal SqlServer(UnitPrices p, Workload w) => p.RdsSqlServerSeLiMultiAzHourly * HoursPerMonth + w.DbStorageGb * p.RdsMultiAzGp3PerGbMonth;
internal sealed record UnitPrices(
decimal WebOnDemandHourly, decimal WebComputeSpHourly, decimal WorkerSpotHourly,
decimal AlbHourly, decimal AlbLcuHourly,
decimal RdsPostgresMultiAzHourly, decimal RdsSqlServerSeLiMultiAzHourly, decimal RdsMultiAzGp3PerGbMonth,
decimal S3StandardPerGbMonth, decimal S3PutPer1000, decimal S3GetPer1000);
internal sealed record Workload(
int WebInstances, decimal AverageLcus, decimal DbStorageGb,
decimal StatementsGb, decimal PutRequests, decimal GetRequests,
int Workers, decimal WorkerHoursPerNight, decimal NightsPerMonth);Output (run in the mcr.microsoft.com/dotnet/sdk:10.0 container on 2026-11-17):
Scenario USD/month USD/year vs A
A All On-Demand, PostgreSQL 555.43 6665.20 0.0%
B Compute SP for web, Spot workers, PostgreSQL 466.08 5592.99 -16.1%
C MISTAKE: SP also sized for the 3-hour batch 966.31 11595.73 +74.0%
D All On-Demand, SQL Server Standard (LI) 1721.97 20663.68 +210.0%Code explained line by line
#:property TreatWarningsAsErrors=true/Nullable=enableโ file-based apps accept MSBuild properties inline; same strictness as every lab project.const decimal HoursPerMonth = 730m;โ the calculator's convention (8,760 hours รท 12). Themsuffix makes it adecimalliteral.new UnitPrices(...)โ every unit price in one record with named arguments, so a reviewer can compare each value with the calculator line by line. Money is alwaysdecimal, neverdouble.new Workload(...)โ the assumptions table as code. ChangeWorkersorWorkerHoursPerNightto see how the batch scenarios move.var shared = Alb(p, w) + S3(p, w);โ costs that are the same in every scenario. CloudFront is 0 because the scenario sits inside the always-free allowance.workerHours = Workers ร hours ร nightsโ 900 instance-hours. On-Demand and Spot bill only these hours.- Scenario C multiplies
(WebInstances + Workers) ร SP rate ร 730โ a Savings Plan commitment is charged for every hour of the month, whether the batch is running or not. That single line is the lesson. var scenarios = new (string Name, decimal Monthly)[] { ... }โ an array of named tuples; easy to extend with your own scenarios.delta = (monthly - baseline) / baselineโ change relative to scenario A.string.Create(gb, $"...")โ formats with the invariant culture, so a UK or EU Windows locale doesn't change the decimal separator or thousands grouping.{delta,8:+0.0%;-0.0%;0.0%}โ a custom format with separate patterns for positive, negative and zero values.static decimal Alb(...)and friends โ local functions, one per service, mirroring the calculator's line items.internal sealed record ...โ immutable data types. In a file-based app, type declarations go after the top-level statements.
Security considerations
- Shared estimates are public. Anyone with the link can open it, and it is stored on AWS public servers. Never put customer names, internal system names that reveal security architecture, real transaction volumes, IP ranges or project code names into descriptions. Use generic labels like
Web tierand fictional volumes. - Real transaction volumes and growth plans of a bank can be commercially sensitive and sometimes inside information. Share estimates through your bank's approved document channels, not public links, unless your information-security team agrees.
- The in-console calculator can see your real usage and discounts; restrict it to FinOps roles in the management account.
โ Learning shortcut โ not for production
This lab uses a public share link because the data is fictional. In a bank, export the estimate to PDF/CSV, strip or replace the link, and store it in the approved repository for the business case. Use the in-console Pricing Calculator (which runs under your organisation's IAM controls) for estimates based on real usage.
โ Learning shortcut โ not for production
The estimate leaves out several real costs (NAT Gateway, cross-AZ traffic, CloudWatch, AWS WAF, AWS Backup, KMS, the support plan, tax). A production estimate must include them, plus a contingency and a growth forecast, and should be reviewed by finance and by someone who has operated a similar workload.
Banking use case
Harbourline's architecture review board template asks for three options for every new service:
- Baseline: everything On-Demand โ the "no commitment" cost, valid from day one.
- Optimised: steady parts on Savings Plans (Compute for EC2/Fargate/Lambda, Database for RDS), interruptible parts on Spot, after 60 days of stable usage.
- Licence alternative: for any Microsoft component, the License Included price and the cost of the open-source alternative or BYOL, with a note from the licensing team.
The board approves the baseline budget, and FinOps later buys commitments for the organisation as a whole rather than per project. The statement portal from this lab would be approved with PostgreSQL. A SQL Server option would need a written justification (for example, a vendor product that only supports SQL Server), sign-off from the licensing team and a migration plan.
Whether a design meets regulatory expectations (operational resilience, outsourcing, data residency) is a separate review. A cost estimate never implies compliance.
Common mistakes
- Pricing only the servers โ it is what you think of first โ data transfer, load balancers and storage are missing. Use the assumptions table as a checklist.
- Covering batch fleets with Savings Plans โ the discount looks attractive โ you pay for idle hours (scenario C).
- Comparing PostgreSQL and SQL Server on different instance sizes โ the comparison becomes meaningless. Keep everything except the engine identical.
- Treating the Spot estimate as a guarantee โ Spot prices and availability change โ present it as a range and keep an On-Demand fallback in the design.
- Losing the share link โ estimates can't be found without it โ paste it into your journal immediately.
- Leaving the Region on the default โ prices differ between Regions โ always set the Region on every service line.
- Putting confidential details in descriptions โ the link is public โ use generic, fictional labels.
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
| Your EC2 line is higher than the sample | EBS volumes added, or Windows selected | Set EBS to 0 or note the extra; check OS is Linux |
| ALB cost far above USD 22 | Traffic inputs produce many LCUs | Lower new connections and processed bytes until about 1 LCU |
| CloudFront shows a charge | Free tier not applied, or a different geography selected | Enable the free-tier option if offered; choose US/Europe; the always-free tier is 1 TB and 10M requests |
| SQL Server option not offered for Multi-AZ | Edition doesn't support it in the calculator, or a very small instance class chosen | Use Standard or Enterprise and db.m6i.large or larger |
| Share link shows an old estimate | Edits are not saved to an existing link | Choose Share again; a new link is created for each save |
| Banner says "Update estimate" on an old link | Prices or service models changed | Choose Update estimate and review lines marked Required inputs |
| C# totals differ by a few cents | Rounding per line vs at the end | Ignore differences below USD 0.05 |
dotnet run StatementPortalEstimate.cs fails | .NET SDK older than 10 | File-based apps need the .NET 10 SDK: dotnet --version |
Cost and cleanup
Cost: USD 0. The public AWS Pricing Calculator is free and needs no account. Nothing is deployed. If you tried the in-console calculator, use only workload estimates (free) โ bill estimates cost USD 2 each after five per month.
Cleanup:
- Delete local exports that you don't want to keep.
- Shared links expire after a year. There is no "delete link" step; just don't publish links that contain anything sensitive.
- Confirm nothing is running (nothing should be):
aws resourcegroupstaggingapi get-resources --tag-filters Key=Project,Values=cea-lab --region us-east-1 --profile learn-sandbox --query "ResourceTagMappingList[].ResourceARN"Exercises
- ๐ข Guided: Complete all steps, fill in the results template and save the share link and CSV export in your journal.
- ๐ก Challenge: Clone the estimate and price the same portal in eu-west-2 (London), with the web tier on AWS Fargate (2 tasks ร 1 vCPU, 2 GB, 730 hours) instead of EC2. Report the difference in each group and explain which part of the change comes from the Region and which from the compute choice.
- ๐ด Live problem: Product management announces that from next quarter statements must be available within 15 minutes of month-end close, instead of overnight. The batch must now process the same work in a 15-minute burst on the first day of each month, plus small daily runs. Rework the batch part of the estimate and write a three-sentence recommendation for the review board.
Assignment
Add docs/cost-estimate-statement-portal.md to your learning-journal repository: the assumptions table, your filled results table, the four scenario totals (from the calculator or the C# program), the share link (the data is fictional), and a 150-word recommendation to the review board.
Solutions
Your results should be close to the sample table. Expect the largest variance on Spot (the calculator uses a historical-average discount, while the sample uses a single snapshot) and on the ALB (depends on the traffic inputs). The recommendation is scenario B with PostgreSQL. Why: the web tier runs 24ร7, so committing to it is low-risk; the batch is interruptible and runs only 3 hours a night, so Spot fits and a Savings Plan does not.
In eu-west-2 most unit prices are a little higher than in us-east-1, so every line rises by a similar percentage โ that part is the Region effect. For Fargate in us-east-1, the list price is USD 0.04048 per vCPU-hour and USD 0.004445 per GB-hour, so 2 tasks ร (1 ร 0.04048 + 2 ร 0.004445) ร 730 โ USD 72.08/month, about half of the 2 ร m7i.large On-Demand line. That is mostly because the tasks are smaller (1 vCPU / 2 GB vs 2 vCPU / 8 GB), not because Fargate is cheaper per unit of capacity. Compare equal sizes before concluding. Compute Savings Plans also cover Fargate. Why: separating the Region effect from the architecture effect is how you avoid wrong conclusions in a review.
A 15-minute monthly burst needs much more parallel capacity for a very short time, so commitments make even less sense. Options: (1) Spot with an On-Demand floor across several instance types (an Auto Scaling group with a mixed instances policy), started shortly before close; (2) AWS Lambda or Fargate tasks fanned out from a queue, which bill per second or per GB-second and need no idle capacity; (3) keep On-Demand as the guaranteed fallback because of the deadline. Recommendation example: "Move statement rendering to queue-driven Fargate tasks with Spot capacity and an On-Demand minimum, so the month-end burst scales out in minutes and costs only for the burst. Do not cover it with a Savings Plan. Because the 15-minute target is customer-facing, we keep an On-Demand floor and an alarm if the queue isn't drained by T+12 minutes." Why: deadlines change the risk appetite for Spot; elasticity, not commitment, is the cost lever for bursts.
Knowledge check
Answer: The public calculator needs no account and uses list prices. The in-console calculator (Billing and Cost Management) needs an account and can apply your discounts, existing Savings Plans and RIs, and even model your whole consolidated bill (bill estimates).
Answer: The commitment is charged every hour of the term. A commitment sized to the batch would be mostly unused for 21 hours a day, so it costs more than On-Demand.
Answer: Anyone with the link; the estimate is stored on AWS public servers. Links created since 31 May 2023 remain valid for one year.
Answer: The SQL Server Standard License Included database. Options: move to PostgreSQL (or Aurora PostgreSQL), use Bring Your Own Media with Software Assurance on RDS if the bank already owns licences, or right-size โ all subject to licensing-team review.
Interview questions
- "How do you produce a cost estimate for a new service?" โ Start from an architecture diagram and an assumptions table, price each component in the Pricing Calculator per Region, group by layer, show On-Demand plus optimised options, list exclusions, and attach the export and link. Revisit after 30โ60 days with Cost Explorer actuals.
- "The estimate said USD 500 but the first bill was USD 800. Where do you look?" โ Cost Explorer grouped by service and usage type: typical culprits are NAT Gateway hours and data processing, cross-AZ traffic, public IPv4 addresses, CloudWatch Logs ingestion, EBS snapshots and forgotten test resources.
- "How would you present licence costs to a business stakeholder?" โ Side by side, same instance size, License Included vs the alternative, monthly and annual, with the migration effort and risk of the alternative and a note that licensing must confirm.
Certification practice
A company wants to estimate the monthly cost of a new architecture before creating any AWS account. Which tool should it use?
A. AWS Cost Explorer B. AWS Budgets C. AWS Pricing Calculator D. AWS Cost and Usage Report
Answer: C. The public Pricing Calculator works without an account and models planned usage. A and D analyse actual past spend in an existing account. B tracks spend against thresholds.
A company with an AWS Organization wants to see how buying an additional Savings Plan would change its next consolidated bill, taking existing Reserved Instances into account. Which option fits best?
A. The public AWS Pricing Calculator B. A bill estimate in the in-console AWS Pricing Calculator C. AWS Trusted Advisor D. The AWS Price List Query API
Answer: B. Bill estimates layer modelled changes onto last month's consolidated usage and existing commitments. A uses list prices only. C gives general checks, not bill modelling. D returns list prices and doesn't include Savings Plans rates.
A web application runs on four instances 24 hours a day. A separate image-processing job runs on 20 instances for 2 hours each night and can be safely restarted. What is the most cost-effective purchasing strategy?
A. Compute Savings Plan covering 24 instances B. Compute Savings Plan covering 4 instances; Spot Instances for the nightly job C. On-Demand for everything D. Spot Instances for the web tier; Compute Savings Plan for the nightly job
Answer: B. Commit to the steady 24ร7 baseline and use Spot for the restartable, short batch. A pays for 20 idle instances 22 hours a day. C misses easy savings. D puts the always-on web tier on interruptible capacity and wastes the commitment.
Completion checklist
- I built a grouped estimate with Web tier, Data and Batch workers groups
- I priced the web tier On-Demand and with a Compute Savings Plan, and the batch three ways
- I compared RDS PostgreSQL with RDS SQL Server Standard License Included on the same instance size
- I saved the share link and exported a CSV, with no confidential information in the descriptions
- I ran the C# program (optional) and explained why scenario C is a mistake
Key takeaways
- A good estimate is an assumptions table plus grouped line items plus clearly stated exclusions.
- Commit (Savings Plans) to steady 24ร7 baselines only; use Spot or serverless for interruptible or bursty work.
- For Microsoft workloads, the licence model can outweigh every other line in the estimate.
- Public calculator links are public: fictional or generic data only.
- Re-check estimates against Cost Explorer actuals after deployment.
Further reading
- What is AWS Pricing Calculator?
- Create and configure an estimate, Using groups, Sharing your estimate
- In-console Pricing Calculator (workload and bill estimates)
- Amazon CloudFront pay-as-you-go pricing and CloudFront flat-rate plans
- Licensing Microsoft SQL Server on Amazon RDS
- Pricing models and Free Tier and billing
- Module 13: Microsoft-to-AWS migration
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.
Pricing models: On-Demand, Savings Plans, Reserved Instances and Spot
Choose the right way to pay for AWS compute and databases for each kind of banking workload, and look up real prices from the console, the CLI and C#.
The Shared Responsibility Model
Draw the line between what AWS secures and what Harbourline Bank still owns, for EC2, RDS, Fargate, Lambda and S3, and collect evidence for your side of the line.