CIDR planning: an IP plan that never overlaps
Plan non-overlapping address ranges for dev, test, prod, two data centres and a partner, size subnets for AWS, and check the plan with a tested C# library (Lab 2.1).
Assumes: .NET 10 LTS ยท CDK v2 (Amazon.CDK.Lib 2.277.0) ยท Terraform 1.14 / AWS provider v6 ยท Region us-east-1. Verified against official sources on 2026-11-17.
Learning objectives
By the end of this lesson you can:
- Explain why overlapping address ranges are the most expensive networking mistake to undo.
- Build an organisation-level IP plan with room for growth: data centres, AWS environments, remote access and subsidiaries.
- Size subnets for AWS (five reserved addresses, AZ growth, container and Lambda ENIs) and keep each tier on an aligned block.
- Express the plan in Terraform (
cidrsubnet), CDK (IpAddresses.Cidr+CidrMask) and check it with the lab'sHarbourline.Networking.CidrPlannerlibrary. - Describe what Amazon VPC IP Address Manager (IPAM) adds in production.
Prerequisites
- Networking refresher โ CIDR arithmetic.
- Amazon VPC fundamentals.
- Tools: .NET 10 SDK, Git (to clone the lab), optionally Terraform.
Why this matters in banking
Banks connect everything: two or more data centres, many AWS accounts, card schemes, payment networks, Open Banking partners, acquired companies and staff VPNs. Routing only works if every address is unique across all of them. When two networks overlap, packets for "10.20.5.7" cannot know which network you mean.
Harbourline Bank learned this with an Open Banking partner, Northgate Payments (fictional). Northgate's integration VPC uses 10.20.0.0/16 โ the same range as Harbourline's dev VPC. A VPC peering request fails immediately, a Transit Gateway attachment would black-hole traffic, and a VPN cannot route to both. The fix is not cheap: either re-address a live network, or put a translation layer (PrivateLink or private NAT) in between. A one-page IP plan, agreed before the first VPC, prevents most of this.
The idea in plain language
An IP plan is a register of who owns which address range, written before anyone builds anything. You start from a large private block, give each big area (data centre, AWS, VPN pool) a slice, then give each environment and VPC a smaller slice inside its area. Every new request is checked against the register.
Analogy: a city's postcode plan. The council assigns district codes first (EC, WC, N, SW), then each district assigns streets. Two streets never share a postcode, and new estates are planned into empty districts โ nobody renumbers Oxford Street after the fact.
The Harbourline register
| Range | Owner / use |
|---|---|
10.0.0.0/14 | On-prem London data centre (LDN-DC1, primary) |
10.4.0.0/14 | On-prem Manchester data centre (MAN-DC2, DR) |
10.16.0.0/12 | All of AWS (10.16.0.0 โ 10.31.255.255) |
10.16.0.0/16 | Network hub / shared services (inspection, egress, DNS resolver endpoints) |
10.17.0.0/16 | prod โ Payments VPC |
10.18.0.0/16 | prod โ Accounts API VPC |
10.19.0.0/16 | test / pre-prod |
10.20.0.0/16 | dev and the learning sandbox (this module's lab VPC) |
10.21.0.0/16 โ 10.31.0.0/16 | Reserved: growth and the EU subsidiary (eu-west-1) |
172.16.0.0/16 | Staff remote-access VPN pool |
100.64.0.0/10 | Never used for workloads (carrier-grade NAT space) |
Why these choices:
- One summarisable AWS block.
10.16.0.0/12lets the data centres send one route to AWS ("everything in 10.16/12 goes over Direct Connect") instead of one per VPC. - Aligned blocks. Each /16 starts on a /16 boundary, each /14 on a /14 boundary, so every range can be written as a single CIDR.
- Reserved space is part of the plan. Eleven unused /16s cost nothing and save a re-addressing project.
- The VPN pool sits outside 10/8. Staff home routers often use 192.168.x.x and sometimes 10.0.x.x; keeping the VPN pool in 172.16/16 avoids clashes with on-prem ranges.
Inside a VPC: tiers and AZs
The lab VPC 10.20.0.0/16 uses /20 subnets in two AZs. Each tier gets an aligned /18 block with room for four AZs:
| Tier | Block | AZ a | AZ b | Room for |
|---|---|---|---|---|
| public | 10.20.0.0/18 | 10.20.0.0/20 | 10.20.16.0/20 | 10.20.32.0/20, 10.20.48.0/20 |
| app (private) | 10.20.64.0/18 | 10.20.64.0/20 | 10.20.80.0/20 | 10.20.96.0/20, 10.20.112.0/20 |
| data (isolated) | 10.20.128.0/18 | 10.20.128.0/20 | 10.20.144.0/20 | 10.20.160.0/20, 10.20.176.0/20 |
| spare | 10.20.192.0/18 | โ | โ | a fourth tier (for example endpoints) |
Two benefits of aligned tier blocks:
- Adding a third AZ never renumbers anything.
10.20.96.0/20is waiting. - One rule covers a tier. A NACL or firewall rule "allow from the app tier" can be
10.20.64.0/18instead of a list.
Sizing rules
- Subtract five per subnet (AWS reservations): a /24 is 251 usable, a /20 is 4,091.
- Count ENIs, not servers. Every Fargate task, every Lambda function's Hyperplane ENI, every interface endpoint (one per AZ), every load balancer node and every RDS instance takes an address. EKS with the VPC CNI gives each pod an address โ plan /19 or larger for pod subnets, or use a secondary CIDR.
- Public subnets can be smaller (they only hold load balancers and NAT gateways); this lab keeps them /20 for simplicity.
- Secondary CIDRs can be added to a VPC later (up to 5 by default, adjustable to 50), but they must come from the plan too, and some ranges cannot be combined (for example, a VPC whose primary block is in 10/8 cannot add a 172.16/12 or 192.168/16 secondary โ VERIFY the exact restriction table).
- IPv6 removes scarcity (each subnet is a /64), but you still need an IPv4 plan for anything that talks to on-premises or partners.
What problem this solves (and IPAM)
A spreadsheet works for one team. A bank needs enforcement. Amazon VPC IP Address Manager (IPAM) holds the plan as pools (for example a top-level pool 10.16.0.0/12 with Regional and environment pools under it), lets accounts request a VPC CIDR from a pool instead of typing one, and alerts on overlaps and unmanaged ranges across the whole Organization.
| IPAM tier | Price (checked 2026-11-17) | What you get |
|---|---|---|
| Free Tier | USD 0 | Single account and Region features |
| Advanced Tier | USD 0.00027 per active IP address per hour | Organization-wide pools, allocation rules, compliance monitoring, BYOIP |
Advanced Tier for 10,000 active addresses โ 10,000 ร 0.00027 ร 730 โ USD 1,971 per month โ a real line item, which is why smaller teams start with a register in Git plus an automated overlap check (exactly what the lab library does).
Limits of a plan: it cannot fix partners or acquisitions that already overlap you. For those, use PrivateLink (the consumer sees an address in its own range) or private NAT. See VPC endpoints and PrivateLink.
Microsoft / Azure comparison
| You know (Microsoft/Azure) | AWS equivalent | What's different |
|---|---|---|
| VNet address space planning | VPC CIDR planning | Same discipline; VPC blocks are /16โ/28 |
| Azure Virtual Network Manager IPAM (preview/GA features) | Amazon VPC IPAM | IPAM pools can be shared to accounts with AWS RAM |
| Hub-and-spoke VNet ranges | Network hub VPC + workload VPCs on Transit Gateway | Summarised routes work the same way |
| Azure reserves 5 addresses per subnet | AWS reserves 5 addresses per subnet | Same count |
| Subnet delegation for services | Dedicated subnets for endpoints, EKS pods | Plan a spare tier block for them |
Architecture
Step-by-step implementation (Lab 2.1 โ the IP-planning exercise)
Gather requirements
Harbourline needs: two data centres (each currently using about 150,000 addresses, growth 50%), AWS with dev, test and prod (prod split into Payments and Accounts), a network hub, a staff VPN for 3,000 users, an EU subsidiary in eu-west-1 with its own dev/test/prod, and room for five more VPCs. Partners must not overlap you.
Choose the area blocks
Each data centre needs about 225,000 addresses with growth: a /14 (262,144) fits. AWS needs at least 9 VPCs of /16 now plus growth: a /12 (16 ร /16) fits with 7 spare. The VPN needs 3,000+: a /20 would do, but a /16 in 172.16/12 keeps it obviously separate.
Fill the worksheet
Copy this table into your notes and fill the right-hand column before you look at the solution.
| Need | Your range |
|---|---|
| LDN-DC1 | |
| MAN-DC2 | |
| AWS supernet | |
| Network hub | |
| prod Payments | |
| prod Accounts | |
| test | |
| dev | |
| EU subsidiary dev / test / prod | |
| Staff VPN |
Split one VPC into tiers
For your dev VPC, choose three tiers and a subnet size. Use aligned tier blocks with room for four AZs.
Check it with code
Clone the lab, run the planner CLI against your VPC, and run --check against the partner range. Then run the unit tests, which encode the rules.
AWS Console instructions
There is nothing to create for the exercise. To see the plan in AWS: Console โ VPC โ IP Address Manager (IPAM) shows IPAM; creating one in Free Tier for a single Region costs nothing, but delete it afterwards to keep the account tidy โ VERIFY current console labels. When you later create a VPC (VPC โ Create VPC), the IPv4 CIDR block field accepts either manual input or IPAM-allocated IPv4 CIDR block.
AWS CLI
# What ranges are already in use in this Region? (start of any planning exercise)
aws ec2 describe-vpcs --profile learn-sandbox --region us-east-1 `
--query "Vpcs[].{Vpc:VpcId,Cidrs:CidrBlockAssociationSet[].CidrBlock}" --output json
# Do any existing subnets overlap the range you want to use? (simple filter on the VPC CIDR)
aws ec2 describe-subnets --profile learn-sandbox --region us-east-1 `
--query "Subnets[?starts_with(CidrBlock, '10.20.')].[SubnetId,CidrBlock,VpcId]" --output table
# Optional, production-style: an IPAM in this Region (Free Tier) with a top-level pool
# aws ec2 create-ipam --operating-regions RegionName=us-east-1 --tier free --profile learn-sandbox --region us-east-1The starts_with filter is only a rough check (it misses ranges such as 10.16.0.0/12 that contain yours). The C# overlap check below is the reliable one.
Infrastructure as Code
// CDK allocates subnet CIDRs in configuration order, AZ by AZ.
// "Reserved" groups take up address space without creating subnets,
// which keeps every tier on its aligned /18 block.
var vpc = new Vpc(this, "Vpc", new VpcProps
{
IpAddresses = IpAddresses.Cidr("10.20.0.0/16"),
MaxAzs = 2,
NatGateways = 0,
SubnetConfiguration =
[
new SubnetConfiguration { Name = "public", SubnetType = SubnetType.PUBLIC, CidrMask = 20 },
new SubnetConfiguration { Name = "public-spare", SubnetType = SubnetType.PUBLIC, CidrMask = 20, Reserved = true },
new SubnetConfiguration { Name = "app", SubnetType = SubnetType.PRIVATE_ISOLATED, CidrMask = 20 },
new SubnetConfiguration { Name = "app-spare", SubnetType = SubnetType.PRIVATE_ISOLATED, CidrMask = 20, Reserved = true },
new SubnetConfiguration { Name = "data", SubnetType = SubnetType.PRIVATE_ISOLATED, CidrMask = 20 },
],
});With two AZs this produces public 10.20.0.0/20 and 10.20.16.0/20, reserves .32 and .48, app 10.20.64.0/20 and 10.20.80.0/20, reserves .96 and .112, and data 10.20.128.0/20 and 10.20.144.0/20. A CDK assertion test in the lab compares these CIDRs with the planner library's output, so the two can never drift apart.
Complete code
The lab ships a tested planning library and a CLI: labs/m02-secure-vpc/src/Harbourline.Networking.CidrPlanner and src/Harbourline.Networking.PlanCli. The main types:
| Type | Purpose |
|---|---|
Ipv4Cidr | Strict parsing (rejects host bits and leading zeros), Contains, Overlaps, Split, Subnet, AwsUsableAddresses |
VpcPlanner.Plan(vpc, tiers, azs, subnetPrefix) | Aligned tier blocks with room for four AZs; throws if the plan does not fit |
OverlapDetector.FindOverlaps(ranges) / Conflicts(proposed, register, allowedParents) | Finds every overlapping pair; checks a new range against the register |
HarbourlineIpRegister.Allocations | The register table above, as code |
A short program using the library (add a project reference to Harbourline.Networking.CidrPlanner):
using Harbourline.Networking.CidrPlanner;
// 1. Plan the test VPC with three AZs and /22 subnets
var plan = VpcPlanner.Plan(Ipv4Cidr.Parse("10.19.0.0/16"), ["public", "app", "data"], ["us-east-1a", "us-east-1b", "us-east-1c"], 22);
foreach (var s in plan.AllSubnets)
{
Console.WriteLine($"{s.Tier,-7} {s.AvailabilityZone,-11} {s.Cidr,-15} usable {s.UsableAddresses}");
}
// 2. Check the partner's range against the bank's register
var partner = NamedRange.Of("northgate-payments", "10.20.0.0/16");
var conflicts = OverlapDetector.Conflicts(partner, HarbourlineIpRegister.Allocations, HarbourlineIpRegister.AwsSupernetName);
Console.WriteLine(conflicts.Count == 0 ? "Partner range is safe to route" : $"Conflict: {string.Join(", ", conflicts.Select(c => c.First.Name))}");Or use the CLI directly:
cd labs/m02-secure-vpc
dotnet run --project src/Harbourline.Networking.PlanCli
dotnet run --project src/Harbourline.Networking.PlanCli -- --vpc 10.19.0.0/16 --prefix 22 --azs us-east-1a,us-east-1b,us-east-1c
dotnet run --project src/Harbourline.Networking.PlanCli -- --check 10.20.0.0/16
dotnet test --solution SecureVpc.slnxExpected output of the first command:
VPC 10.20.0.0/16 (65,536 addresses)
Tier Block AZ Subnet Usable
public 10.20.0.0/18 us-east-1a 10.20.0.0/20 4,091
public 10.20.0.0/18 us-east-1b 10.20.16.0/20 4,091
app 10.20.64.0/18 us-east-1a 10.20.64.0/20 4,091
app 10.20.64.0/18 us-east-1b 10.20.80.0/20 4,091
data 10.20.128.0/18 us-east-1a 10.20.128.0/20 4,091
data 10.20.128.0/18 us-east-1b 10.20.144.0/20 4,091
spare 10.20.192.0/18
Subnet overlap check: none
Register check: no conflicts with Harbourline allocationsAnd --check 10.20.0.0/16 prints CONFLICT: 10.20.0.0/16 overlaps 1 allocated range(s): aws-dev-sandbox with exit code 2 โ a pipeline can fail on it.
Code explained line by line
Ipv4Cidr.Parse("10.19.0.0/16")โ strict parsing:10.19.1.0/16throws with a hint (did you mean 10.19.0.0/16?) instead of silently truncating, because a typo in an IP plan is a production incident later.VpcPlanner.Plan(vpc, tiers, azs, 22)โ the tier block is the subnet prefix minus 2 (/20here), giving four AZ slots per tier. With a /16 VPC that is 16 tier blocks, so the three tiers fit with 13 spare.s.UsableAddressesโ the subnet size minus AWS's five reserved addresses (1,019 for a /22).NamedRange.Of("northgate-payments", "10.20.0.0/16")โ every range in the register has a human name, so error messages say who you collide with.OverlapDetector.Conflicts(partner, register, AwsSupernetName)โ checks the proposal against every allocation; theallowedParentsargument ignores the AWS supernet containing a new VPC (that is intended), but would still flag a partner range overlapping it if you did not list it.- Inside the library,
FindOverlapssorts ranges by start address and stops scanning once a later range starts after the current one ends โ the same "sweep" idea used in calendar clash detection. - The tests in
tests/Harbourline.Networking.CidrPlanner.Testscover parsing edge cases, the exact lab plan, "adding an AZ never renumbers", partner conflicts and the bank register itself.
Security considerations
- The IP register is sensitive: it maps your internal estate. Keep it in a private repository with reviewed changes.
- Never rely on address ranges as authentication ("anything from 10.0.0.0/8 is trusted"). Ranges are for routing; identity (IAM, mTLS) is for trust.
- Avoid
100.64.0.0/10and ranges used by AWS services (for example172.17.0.0/16is Docker's default bridge on build agents) โ VERIFY your tooling's defaults. - Changes to the plan affect firewall rules on-premises; route every change through network and security change management.
โ Learning shortcut โ not for production
The lab keeps the register as C# code and checks overlaps in unit tests. In a bank the register lives in Amazon VPC IPAM (or the bank's IPAM product), accounts receive CIDRs by allocation from shared pools, and an automated check runs in the IaC pipeline before any VPC, VPN or partner connection is created.
Banking use case
Harbourline's network team owns an IPAM top-level pool 10.16.0.0/12, split into Regional pools (us-east-1, eu-west-2, eu-west-1) and then environment pools. The Payments team's CDK app asks for "a /16 from the prod-eu-west-2 pool" instead of typing a CIDR. When Northgate Payments' onboarding form arrives with 10.20.0.0/16, the pipeline's overlap check fails, and the architecture decision is made early: expose Harbourline's partner API to Northgate through a PrivateLink endpoint service, which works regardless of overlapping ranges.
Common mistakes
- Every VPC gets
10.0.0.0/16(the wizard default) โ no two VPCs can ever be connected โ assign from the plan. - Sizing subnets by number of servers โ Lambda, Fargate and EKS exhaust addresses โ count ENIs and pods.
- Unaligned blocks (
10.20.10.0/20is invalid;10.20.8.0/21"works" but breaks summarisation) โ usecidrsubnetor the planner. - No reserved space โ the fourth AZ or a new tier forces re-addressing โ reserve slots per tier.
- Forgetting on-premises and VPN ranges โ AWS plan is tidy but collides with the data centre โ one register for the whole bank.
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
InvalidSubnet.Range when creating a subnet | Subnet not inside the VPC CIDR, or host bits set | Recalculate with cidrsubnet or the planner |
InvalidSubnet.Conflict | Overlaps an existing subnet in the VPC | List subnets; pick the next free block |
| VPC peering request fails with overlapping CIDRs | The two VPCs share addresses | Re-address one side, or use PrivateLink |
InsufficientFreeAddressesInSubnet (Lambda/ECS) | Subnet too small for ENIs | Larger subnets or a secondary CIDR |
| Planner throws "only N tier blocks" | Subnet prefix too large for the number of tiers | Use smaller subnets or fewer tiers |
Cost and cleanup
Cost: USD 0. Planning, the CLI and the tests run locally. If you created an IPAM in Free Tier, delete it:
aws ec2 describe-ipams --profile learn-sandbox --region us-east-1 --query "Ipams[].IpamId"
aws ec2 delete-ipam --ipam-id ipam-0123456789abcdef0 --cascade --profile learn-sandbox --region us-east-1
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 the worksheet, then run the planner CLI for your dev and test VPCs and
--checkyour EU subsidiary ranges. - ๐ก Challenge: Add a fourth tier
endpointswith /24 subnets to the dev VPC without changing the existing /20 tiers. (Hint: the spare /18 can be split differently from the others.) Write a unit test for it. - ๐ด Live problem: Harbourline acquires a small fintech whose AWS estate uses
10.16.0.0/16,10.17.0.0/16and172.31.0.0/16(default VPCs). The business wants their customer data API reachable from Harbourline's Payments VPC within a month. Propose two options with trade-offs.
Assignment
Commit docs/ip-plan.md with your filled worksheet, a Mermaid diagram of your plan and the output of PlanCli for two VPCs. Add a CI step (GitHub Actions) that runs dotnet test on the planner so a pull request that breaks the register fails.
Solutions
| Need | Range |
|---|---|
| LDN-DC1 | 10.0.0.0/14 |
| MAN-DC2 | 10.4.0.0/14 |
| AWS supernet | 10.16.0.0/12 |
| Network hub | 10.16.0.0/16 |
| prod Payments | 10.17.0.0/16 |
| prod Accounts | 10.18.0.0/16 |
| test | 10.19.0.0/16 |
| dev | 10.20.0.0/16 |
| EU dev / test / prod | 10.21.0.0/16, 10.22.0.0/16, 10.23.0.0/16 |
| Staff VPN | 172.16.0.0/16 |
Why: /14s fit 225,000 addresses each with headroom; the AWS /12 is one summary route for the data centres; EU ranges come from the reserved space, so nothing existing changes. Other valid answers exist โ what matters is no overlaps, aligned blocks and documented reserve space.
Use the planner for the three /20 tiers, then split the spare block: Ipv4Cidr.Parse("10.20.192.0/18").Subnet(24, i) gives 10.20.192.0/24, 10.20.193.0/24, โฆ Test: assert the endpoint subnets are inside plan.SpareBlocks[0] and OverlapDetector.FindOverlaps over all subnets is empty. Why: aligned spare blocks can be cut differently later without touching existing tiers.
Option 1 โ PrivateLink: the fintech puts an NLB in front of its API and publishes an endpoint service; Harbourline creates an interface endpoint in the Payments VPC. Works despite the overlap (each side keeps its own addresses), one-directional, fast to deliver; cost โ USD 0.01/hour per endpoint per AZ plus NLB. Option 2 โ re-address: move the fintech's workloads into new VPCs allocated from Harbourline's plan (for example 10.24.0.0/16), then connect via Transit Gateway. Correct long-term, slow and risky. A realistic answer: PrivateLink now, re-address as part of the integration programme, and delete the default VPCs immediately. Note that any data sharing needs data-protection review (new controller/processor relationships).
Knowledge check
Answer: On-premises routers and firewalls can use a single summary route and rule for all of AWS, and the register stays simple. Scattered ranges mean a route and a firewall change for every new VPC.
Answer: 1,024 โ 5 = 1,019.
Answer: No. 10.20.128.0/17 is inside 10.20.0.0/16, so they overlap; peering requires non-overlapping CIDRs.
Answer: It allocates address space in the CIDR sequence but creates no subnets, keeping room for later subnets (and, in the lab, keeping each tier on an aligned block).
Interview questions
- "How would you plan IP addressing for a bank moving to AWS?" โ One register for on-prem, cloud, VPN and partners; a summarisable AWS supernet; per-environment /16s; aligned tier blocks with AZ growth; IPAM pools with allocation rules; automated overlap checks in the pipeline.
- "A partner overlaps your ranges. What now?" โ PrivateLink for service-level access (works with overlaps), or private NAT; re-addressing only if the relationship is long-term and deep.
- "Why do serverless and container platforms change subnet sizing?" โ Each task, pod or function ENI consumes an address; bursts can exhaust small subnets.
Certification practice
A company's VPC is 10.30.0.0/16. It needs three tiers, /20 subnets, two AZs now and possibly four later, without renumbering. Which allocation meets this?
A. Allocate /20s sequentially: tier 1 gets 10.30.0.0/20 and 10.30.16.0/20, tier 2 gets 10.30.32.0/20 and 10.30.48.0/20, tier 3 the next two B. Give each tier an aligned /18 (10.30.0.0/18, 10.30.64.0/18, 10.30.128.0/18) and use the first two /20s in each C. Use one /16 subnet per tier D. Use /28 subnets so more can be created later
Answer: B. Each /18 holds four /20s, so AZs 3 and 4 fit without moving anything. A leaves no gap for new AZs in each tier. C is impossible (a subnet cannot equal the whole VPC per tier, and there would be no room for three). D is far too small for most workloads.
A bank must let a partner's application call one internal API. Both companies use 10.0.0.0/16 for their VPCs. Traffic must stay private and the partner must not reach anything else. What should the architect do?
A. Create a VPC peering connection and restrict it with security groups B. Attach both VPCs to a shared Transit Gateway C. Expose the API through an NLB-backed VPC endpoint service and have the partner create an interface endpoint D. Publish the API on a public ALB with an IP allow list
Answer: C. PrivateLink works with overlapping CIDRs and exposes only that service. A and B fail because the CIDRs overlap (and would expose more than one API). D sends traffic over the internet.
Completion checklist
- I completed the worksheet and compared it with the solution
- I ran
PlanClifor two VPCs and the--checkmode - I ran the planner unit tests (
dotnet test --solution SecureVpc.slnx) - I can explain aligned tier blocks and why they help routing and NACLs
- I deleted any IPAM I created
Key takeaways
- Overlapping ranges are expensive to fix; plan the whole bank's address space first.
- Use a summarisable AWS supernet and aligned blocks at every level.
- Size subnets by ENIs (containers, Lambda, endpoints), and subtract five.
- Keep the plan in a reviewed register; check every new range automatically.
- When overlaps are unavoidable, PrivateLink beats re-addressing in the short term.
Further reading
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.
Amazon VPC fundamentals
Understand what a VPC is, what AWS puts in the default VPC, how VPC DNS works, and why banks delete default VPCs and turn on VPC Block Public Access.
Public, private and isolated subnets
Build a three-tier subnet layout across two Availability Zones, explain exactly what makes a subnet public, private or isolated, and audit any VPC's subnets with a C# tool.