Amazon AWS SAA-C03: Direct Connect Resiliency

AWS Direct Connect resiliency depends on more than having two circuits. For Amazon AWS SAA-C03, the design question is whether the location, device, provider path, BGP policy, virtual-interface configuration, and failover process are independent enough for traffic to keep moving when a component fails.

Within AWS architecture, Direct Connect is the hybrid-network path between on-premises or colocated environments and AWS. BGP path selection determines which redundant path actually carries traffic, so a visually symmetric network diagram is not evidence of symmetric failover.

AWS provides the Direct Connect Resiliency Toolkit to help teams design and test this path around explicit resilience objectives.

Maximum resiliency protects against a full Direct Connect location failure

The maximum-resiliency model uses separate connections that terminate on separate devices in more than one Direct Connect location. AWS associates this design with the requirements needed to qualify for the 99.99% Direct Connect SLA, subject to the full SLA conditions.

The important architectural idea is independent failure domains. Two circuits terminating on the same physical device or in the same location do not protect against every failure the organization may care about.

Maximum resiliency is appropriate for critical workloads whose hybrid path must tolerate a complete Direct Connect location failure. Provider diversity should be verified rather than assumed. Two connections ordered from different carriers can still share conduit, meet-me room, or upstream infrastructure. The business should decide which physical failures matter and ask providers how independently the paths are engineered where that information is available.

High resiliency uses multiple locations with fewer connections

The high-resiliency model uses two single connections to multiple Direct Connect locations. AWS associates this design with the requirements for the 99.9% SLA, again subject to the service’s SLA terms. This protects against common fiber, device, and location failures with less infrastructure than the maximum-resiliency design.

The organization should choose the model based on business impact and recovery expectations rather than treating maximum redundancy as automatically correct for every workload.

Customer-edge devices also need redundancy. Two AWS-side connections can terminate on one on-premises router, firewall, or power circuit and still share a single local failure domain. Maximum-resiliency design should be mirrored through the customer side until traffic reaches an infrastructure layer that is intentionally shared.

Bandwidth asymmetry can create a hidden failover problem. The primary path may carry several gigabits while the surviving path has less capacity. During a test, basic connectivity remains healthy but queues and application latency rise sharply. Resiliency planning should size the secondary path for the traffic the business expects to preserve during failure.

Development and test resiliency stays inside one location

AWS also documents a development/test model that uses separate connections terminating on separate devices in one Direct Connect location. This protects against device failure but not a complete location failure. That can be appropriate for non-critical workloads where the cost and complexity of multiple locations is not justified. The design should be labeled honestly so a development pattern is not later mistaken for production-level regional resiliency.

BGP policy determines which redundant path actually carries traffic

Direct Connect circuits carry virtual interfaces that establish BGP sessions for public, private, or transit connectivity. A redundant physical design only helps if route advertisements and local preference cause traffic to move to the intended surviving virtual interface.

BGP attributes, route filtering, transit-gateway design, and on-premises routing policy can all change which path is preferred. Route domains add route propagation and segmentation decisions to that BGP design, which can change whether a surviving Direct Connect path reaches the VPCs the application actually needs.

BFD can reduce failure-detection time in some routing designs, but faster detection should be tested with device and provider behavior. Aggressive timers that are unstable under normal packet loss can create route flaps worse than the failure they are meant to detect. Routing convergence should be measured end to end rather than inferred from one protocol timer.

Route advertisement policy should be documented before the test. If both links advertise the same prefixes, local preference and AS-path behavior determine which path carries traffic normally and which becomes backup. For transit virtual interfaces, Transit Gateway route propagation and segmentation add another layer that can alter the expected path.

The Direct Connect failover test proves route movement without a cable pull

AWS provides a failover test that places selected BGP peering sessions into the DOWN state for a configured period. The team can then send traffic and verify that it moves over the redundant virtual interface. The test works with public, private, transit, and hosted virtual interfaces. It can be stopped early, and AWS restores the BGP session when the test ends.

This is valuable because it tests the actual routing behavior without requiring a physical cable pull. Regular tests should vary which peering is brought down. Testing only one link proves only one direction of failover. A complete exercise rotates through the connections and validates that each surviving topology meets the requirement, including return traffic and stateful firewall behavior.

Application-path validation has to go beyond BGP state

A successful routing-table change does not prove the application path is healthy. The test should include DNS, firewall state, MTU, asymmetric routing, bandwidth, and real application traffic across the surviving circuit. If the backup path has lower bandwidth, the application may technically remain reachable while service-level objectives fail under production load.

The test should therefore measure both connectivity and performance. Monitoring should track BGP session state, connection state, bandwidth utilization, packet loss where observable, and application latency. A link can remain “up” while becoming too degraded to meet the workload SLO. Operational policy should define when to prefer the alternate connection before complete failure.

Failover testing should avoid active maintenance windows

AWS warns against using the failover-test feature during a Direct Connect maintenance period because the BGP session can be restored prematurely during or after the maintenance. Change calendars and AWS maintenance notifications should therefore be checked before the exercise begins.

Resilience tests are most useful when the environment is otherwise stable enough that operators can attribute the result to the failure they intentionally introduced.

Maintenance events are another reason to rehearse. AWS or a provider can schedule work that affects one connection or device. If the redundant path has been tested recently, maintenance becomes a controlled capacity event rather than an unplanned experiment with production traffic.

VPN can provide an independent backup path when the routing policy supports it. Some organizations combine Direct Connect with Site-to-Site VPN so a separate internet-based path exists if Direct Connect connectivity is unavailable. The routing design determines whether VPN is active/standby or used for selected traffic. When Site-to-Site VPN is the independent backup path, IPsec Site-to-Site VPN design determines tunnel establishment, encryption, routing, and failure behavior on the internet-based path that must carry traffic after Direct Connect is unavailable.

The backup path should be sized and tested for the traffic it is expected to carry. A VPN that has never handled production volume is only a theoretical fallback.

Direct Connect encryption requirements should also be addressed where relevant. MACsec is available on supported dedicated connections and can protect traffic over the physical Direct Connect link, while application- or VPN-layer encryption may still be used depending on the security model. Resilience and confidentiality should be designed together rather than assuming a private circuit automatically satisfies every data-protection requirement.

A failure matrix should name the surviving path and expected operator action

Teams should document the failures they expect the design to survive: one BGP session, one customer router, one Direct Connect device, one circuit, one provider path, one location, or the entire hybrid connection. Each row in that matrix should map to the path that remains, the expected traffic shift, the monitoring signal, and the operator action.

Direct Connect resiliency becomes trustworthy when the routing and physical design have been exercised under those failures, not merely when the connection count matches a diagram template.

Direct Connect Gateway and Transit Gateway designs should be included in failover tests. A circuit can recover at BGP while route propagation, association, or segmentation prevents a destination VPC from receiving the expected path. Hybrid tests should therefore include representative VPC routes, not only one reachability target.

Runbooks must cover route convergence, inventory, and provider escalation

The connection inventory should also record LOA/CFA information, provider, location, device, VLAN, virtual-interface ownership, and the business services that depend on each path. Hybrid connectivity is difficult to restore quickly when technical and contractual details live only in one engineer’s notes.

Route convergence time should be measured during every failover exercise. The business cares about the period during which application traffic is interrupted or degraded, not merely the moment the BGP session changes state. Timers, provider behavior, firewall session handling, DNS dependencies, and application retry logic all influence the observed recovery time.

Asymmetric routing should be checked explicitly, especially when multiple connections, firewalls, and Transit Gateway paths are involved. One direction can move to the backup circuit while return traffic still prefers the primary path, creating stateful-firewall or troubleshooting problems even though both routes look individually valid.

Finally, ownership should span both AWS and the carrier side. Incident runbooks should include provider escalation contacts, circuit IDs, demarcation details, and the AWS resources associated with each connection. A technically redundant design can still have a long outage if nobody can identify or escalate the failed circuit quickly.

Leave a Reply

How It Works

img
Step 1. Choose Exam
on ExamLabs
Download IT Exams Questions & Answers
img
Step 2. Open Exam with
Avanset Exam Simulator
Press here to download VCE Exam Simulator that simulates real exam environment
img
Step 3. Study
& Pass
IT Exams Anywhere, Anytime!