Microsoft AZ-104: Azure VPN Gateway Design
A VPN Gateway design succeeds when the required networks exchange routes securely and recover predictably from tunnel, device or gateway failures. Picking a gateway SKU is only one part of planning authentication, routing, resilience and operations. A working VPN tunnel is only the beginning of connectivity. Engineers must validate advertised routes, overlapping address ranges, authentication choices, gateway redundancy, and the behavior users see when one peer or circuit fails.
On this page
- Choose the connection pattern before the gateway
- Plan address space and effective routes first
- Select gateway redundancy to match outage objectives
- Establish security policy and cryptographic ownership
- Diagnose tunnels, BGP and application traffic separately
- Plan coexistence with ExpressRoute and other hubs
- Prove recovery under a controlled network test
Choose the connection pattern before the gateway
Azure VPN Gateway supports site-to-site connectivity for on-premises networks and supported virtual-network patterns, as well as point-to-site connections for individual clients. These have different authentication and routing requirements. A branch office linking a data center LAN to Azure is not equivalent to a remote employee connecting from a laptop. An effective Microsoft AZ-104 VPN design exercises route propagation, encryption settings, failover and the return path under failure conditions.
Record source prefixes, intended destinations, protected workloads and the users or devices allowed to use each path. Define which applications need bidirectional connectivity and which should only initiate outbound connections. The topology should distinguish gateway, local network gateway representations, actual on-premises VPN devices, VNet/subnet boundaries and any central firewall.
Microsoft’s VPN Gateway design guidance describes supported site-to-site topologies, multi-site connections and coexistence with ExpressRoute in appropriate configurations. Do not design an unsupported network around a generic diagram that omits gateway type, connection capabilities or the necessary gateway subnet.
Plan address space and effective routes first
Overlapping IP address ranges are common in mergers and multi-branch estates. A tunnel can establish successfully while routes cannot uniquely identify the remote network. Inventory VNet address spaces and all on-premises prefixes before configuration, including prefixes propagated by transit or routing appliances. Decide how overlap will be handled, potentially with supported NAT or renumbering where appropriate.
Site-to-site designs typically use static or BGP-learned prefixes according to supported connection settings. BGP can simplify changing networks and failover, but creates a dependency on correct peer addresses, autonomous systems, advertised routes and filtering. A route that is advertised is not proof it is preferred or reachable from every application subnet.
Trace a test packet from workload source through effective VNet routes, VPN gateway, remote device and destination, then trace the return path. Asymmetric forwarding can cause stateful inspection devices to reject otherwise valid traffic. Examine NSGs, local firewalls, user-defined routes and DNS resolution before attempting to repair a routing issue by widening network-security rules.
Select gateway redundancy to match outage objectives
An active-standby VPN Gateway configuration can fail over between instances under supported conditions, but a failover can interrupt sessions. Active-active mode supports tunnels from both gateway instances where configured correctly with compatible peer devices. Microsoft’s active-active documentation emphasizes that a connection to only one active instance defeats the intended redundancy.
A highly resilient site-to-site topology can use independent on-premises VPN devices and multiple tunnels, avoiding one physical device or WAN uplink as a common failure point. However, a small branch firewall may not support all such combinations. Verify the vendor’s compatible tunnel, encryption and routing capabilities before presenting a four-tunnel mesh as an automatic option.
In supported Azure regions and gateway SKUs, zone-redundant gateway deployment can reduce the impact of a single availability-zone event. Zone support does not remove dependence on ISP diversity, route correctness or the enterprise’s own gateway equipment. Capacity planning should account for traffic during degraded operation, when fewer usable tunnels may carry the complete workload.
Establish security policy and cryptographic ownership
For a remote workforce using point-to-site VPN, document how clients receive configuration and certificates, which identity group authorizes a connection, how credentials are revoked, and whether split tunneling is permitted. Split tunneling can improve efficiency for selected traffic but also changes which destinations traverse enterprise inspection and DNS. Test an ordinary staff device and a device whose access has just been revoked. A configuration file that still contains old server information can make revocation appear ineffective even when the intended policy was updated elsewhere.
A VPN tunnel encrypts traffic between endpoints, but it does not guarantee the endpoints are authorized to reach every workload behind the tunnel. Apply appropriate segmentation and firewall policy, keep permitted prefixes narrow and protect the control plane of both Azure and on-premises devices. Manage shared keys, certificates and client authentication with a documented rotation and emergency process.
For site-to-site connections, agree on compatible IPsec/IKE parameters and supported policy combinations. Weak or mismatched proposals can prevent negotiation or create unacceptable security posture. For point-to-site, choose supported client authentication, certificate or Microsoft Entra integration as applicable and plan device onboarding, revocation and policy distribution.
Log connection and configuration changes with enough context to distinguish expected rekeys from repeated authentication failures. In a shared platform, an engineer should not need to expose a VPN pre-shared key to application owners merely to troubleshoot a packet drop. Define support roles and the minimal diagnostics they need.
Diagnose tunnels, BGP and application traffic separately
As a practical diagnostic example, imagine a branch file server at a private address that responds from one Azure subnet but times out from another. Both branches show a green connection indicator. Compare the effective routes for the two Azure subnets, the remote device’s advertised return prefixes and any route filters. If the second subnet’s prefix is missing on-premises, opening TCP port 445 in its NSG cannot fix the return path. If DNS resolves the server to a public address instead, investigate the resolver chain. A single precise packet path can rule out several unrelated theories.
Troubleshooting works best as a sequence. First confirm the VPN gateway resource and connection state. Next inspect tunnel negotiation, routing advertisements and learned prefixes, then verify the actual application’s network path. A connected tunnel might carry no useful traffic because the remote gateway has no route to the target VNet or the application points at the wrong address.
In active-active mode, each instance has its own BGP peer address. Microsoft’s BGP troubleshooting guidance discusses the significance of independent sessions and asymmetric routing. A design can look healthy because one peer has learned routes while the second path remains incorrectly configured, only failing visibly during maintenance.
Measure from both sides. Confirm DNS returns the expected private address, effective routes include the target prefix, the security policy allows the specific flow and the application responds. A blocked TCP connection, a DNS miss, a route black hole and an invalid user token are four separate incidents. Avoid resetting a gateway just because every remote-access issue is labeled “VPN down.”
Plan coexistence with ExpressRoute and other hubs
Some enterprises design resilient ExpressRoute connectivity as the preferred private path and a site-to-site VPN as an emergency alternate. Supported coexistence topologies require appropriate gateway types, route preferences and network configuration. Do not equate a redundant gateway instance with a truly independent alternate WAN provider.
A backup VPN may have lower throughput or higher latency, so decide in advance which services remain essential during failover. Validate route advertisements to avoid accidentally preferring the internet backup during normal conditions. If the Azure environment uses a hub-and-spoke or managed Virtual WAN design, gateway connectivity and route propagation may be governed by a different control-plane model.
Document which team owns each route: network provider, central cloud platform, site appliance or workload VNet. If a route fails between these boundaries, the responders need a shared path trace rather than several teams each reporting their own device as healthy.
Prove recovery under a controlled network test
Include a route-change rollback sheet in each gateway runbook: expected peer addresses, accepted prefixes, preferred next hop, approved IPsec algorithms, shared-key rotation owner and a pre-change configuration export. During a drill, avoid resetting every peer when one tunnel appears bad; preserve diagnostic state long enough to identify the failing boundary. After the link returns, check that route preference normalizes and that monitoring no longer shows asymmetric drops. Successful reconnection is not enough if the workload remains pinned to a temporary, lower-capacity path.
A resilience test should intentionally remove a selected tunnel or peer in a permitted test environment while retaining traffic measurements. Observe packet loss, BGP reconvergence, DNS and application response. For active-active operation, check that the surviving path carries representative traffic, not simply that it continues to display “connected.”
Record baseline throughput and the maximum acceptable failover interruption for critical applications. Test the reverse transition as well: restoring a route may reintroduce asymmetry or reset long-lived sessions. Schedule controlled maintenance with a rollback and stakeholder communication plan before broad changes to production routing.
A reliable VPN design documents the route a permitted packet takes during normal operation, the alternate route under failure, and the security controls enforced along both paths. Exercise a supported peer failure and measure actual application interruption rather than accepting a green tunnel indicator. During restoration, confirm that preferred routes return and that no unexpected asymmetric flow remains.