zero gravitySOLUTIONS
← All insightsCLOUD & OPERATIONS / ARCHITECTURE DECISION GUIDE

Private AWS networking without an unexplained egress bill.

QUICK ANSWER

Choose endpoints and NAT from a destination-by-destination route matrix. Verify DNS and access as well as connectivity, and explain costs using endpoint-hours, gateway-hours, processed data, and transfer paths.

Inside this article 6 sections

Choose VPC endpoints or a NAT gateway by mapping each outbound destination, the route and DNS behavior it requires, and the costs attached to that path. Endpoints can keep supported AWS service traffic on a private path; a NAT gateway can provide outbound access for destinations that still require public addressing.

A private subnet is not a complete egress design. The application may need S3, Secrets Manager, a container registry, logging, and an external payment API, each with different endpoint support and access controls.

Build a destination ledger before changing routes

For an illustrative document-processing service, list startup traffic and steady-state traffic separately. A task may need to retrieve an image and secret before it starts, then read customer objects, publish logs, and call an external conversion provider. A network change can break startup even when the running application still appears healthy.

Record the destination hostname, service, Region, account owner, expected data volume, and required identity. Include package downloads and operational tooling if they occur in the runtime environment. An undocumented outbound dependency often appears only when a new task launches or an incident repair begins.

Map the selected path for each destination. Avoid a single label such as “private AWS traffic” that hides whether the request uses a gateway endpoint, an interface endpoint, or a public endpoint through NAT.

Original route matrix · illustrative processing workload
DestinationCandidate pathVerification point
Same-Region S3 objectsGateway endpoint where appropriateRoute table and bucket policy agree
Secrets Manager APISupported interface endpointPrivate DNS, security group, and endpoint policy
External conversion APIControlled public egress, such as NATDestination connectivity and outbound identity
Container startup dependenciesService-specific endpoint or egress setA fresh task starts without hidden fallback routes
The matrix is a design worksheet. Verify endpoint availability and required companion services for the actual Region and runtime.

Distinguish gateway endpoints from interface endpoints

Gateway endpoints provide a route-table-based path for supported services such as S3 and DynamoDB. Interface endpoints use network interfaces and AWS PrivateLink for supported service connections. Their routing, DNS, security, and charging models differ, so “add an endpoint” is not a complete implementation instruction.

For interface endpoints, choose the required subnets and Availability Zones, configure security groups, and decide how private DNS should resolve the service hostname. A private IP address alone does not establish that the client uses the endpoint; the application's resolver and hostname choice matter.

Check cross-network requirements. A path suitable for resources inside one VPC may not be the same path needed by on-premises clients or another VPC. Centralizing endpoints can add routing and DNS dependencies that should be weighed against the operational and cost benefits.

Treat DNS as part of the route

When private DNS is enabled for an interface endpoint, clients using the normal service name may resolve to private addresses. Verify this from the actual workload environment. A developer laptop or bastion host can use a different resolver and produce a misleading connectivity result.

S3 supports endpoint arrangements with specific DNS behavior, including cases involving inbound Resolver paths. Do not copy an S3 interface-endpoint setting into a design that intends VPC traffic to use a gateway endpoint without checking where each request will go.

Test both name resolution and an authenticated service request. A successful DNS lookup proves only that an address was returned. Security groups, endpoint policies, resource policies, and IAM permissions can still deny the intended operation.

Explain cost with path-specific quantities

Build a cost worksheet from endpoint-hours, NAT gateway-hours, processed data, transfer paths, and Availability Zone placement. Apply current regional rates separately. Avoid comparing only per-gigabyte rates while ignoring the number of endpoint interfaces or continuously provisioned gateways.

Suppose an illustrative workload sends 600 GB to supported AWS services and 100 GB to an external provider in a reporting period. Moving the 600 GB to an appropriate endpoint path changes which service processes that traffic; it does not automatically remove the NAT gateway if the 100 GB still needs it.

Likewise, concentrating egress in one Availability Zone can change transfer paths and failure exposure. Record the intended route during normal operation and during an Availability Zone disruption. A cheaper-looking normal path may need a more complex recovery procedure.

The cloud cost allocation guide helps distinguish directly attributable traffic from shared network overhead. Keep unclassified egress visible rather than assigning it to customers through an unexplained formula.

Review access policies alongside connectivity

An endpoint provides a network path; it does not replace application authorization. Limit endpoint policies and resource policies where useful, and preserve the task's least-privilege IAM role. Confirm that the account and Region in the request are the intended ones.

Do not remove the public egress path until fresh tasks, credential refresh, logging, and recovery tools have been tested. If an external API remains necessary, give that path a named owner and observable usage. “No NAT” is not an architecture objective unless it serves a concrete security, cost, or operating requirement.

Validate a fresh runtime

Launch a new test task, retrieve its secret, read a scoped object, emit telemetry, and call the approved external dependency. Then disable the fallback path in the test environment and confirm that intended endpoint traffic still works. Record failures by destination and policy layer.

The AWS account boundary guide clarifies ownership across those services. Architecture design can produce a route matrix, access model, and cost worksheet that explain the egress bill and survive a fresh deployment.

FROM INSIGHT TO ACTIONSoftware architecture consulting

Make private networking and its bill understandable.

We can design endpoint and NAT routes, validate fresh-runtime dependencies, and build a path-specific cost worksheet.

Explore architecture consultingDiscuss your architecture

Start with your goals.
We’ll work through the next step together.