If there is no second data center, the DR site is often Azure: servers are replicated there and powered on there in a disaster. Azure Site Recovery (ASR) protects on-premises Hyper-V, VMware vSphere and physical servers with this model; in normal times you pay only for replica disks, and virtual machines are created at failover. It looks simple on paper. In practice, what makes it hard isn’t replication, but networking, DNS and drill discipline. Below we explain the architecture, policy, network design, recovery plans and test failover in the order of implementation.
Azure Site Recovery components
The components vary by the type of resource being protected. What they have in common is that all configuration is kept in a Recovery Services vault in Azure.
| Source | On-premises component | Replication method |
|---|---|---|
| Hyper-V (no VMM) | ASR Provider and Recovery Services agent on each host | Hyper-V Replica infrastructure |
| Hyper-V (with VMM) | Provider in VMM, agent on hosts | Hyper-V Replica infrastructure |
| VMware vSphere | Replication appliance, Mobility service on VMs | Continuous block replication |
| Physical server | Replication appliance, Mobility service on the server | Continuous block replication |
In VMware and physical server infrastructure, the replication appliance receives data from the sources, writes it to cache, and compresses, encrypts and sends it to Azure. On the Azure side, data is first written to a cache storage account and from there processed into the target managed disks. Until failover, there is no virtual machine running in Azure.
For VMware and physical servers, Microsoft recommends the modernized appliance-based architecture instead of the old “configuration server” architecture. Check in the Azure Site Recovery documentation which architecture is supported for new deployments. Starting from an old setup guide creates migration work later.
Supported sources and requirements
Before moving on to setup, check the support matrix for each server one by one. Common sticking points:
- Operating system and kernel version: On Linux, the Mobility service supports specific kernel versions; a routine kernel update can silently stop replication. Add an ASR check to the patch process.
- Disk size and count: Disks that don’t fit Azure’s managed disk and VM size limits can’t be protected.
- High data change rate (churn): There is a supported upper limit of data change for each disk and VM. On write-heavy database servers, this limit determines the target disk type (Standard or Premium SSD).
- Shared disks: Shared VHDX/RDM disks used in classic guest cluster setups need a separate support check.
- Outbound connectivity: The hosts or appliance must be able to reach Azure endpoints over HTTPS (443); if there is a proxy, define the exceptions. A firewall doing SSL inspection can also break replication.
To do these checks in bulk, Microsoft offers the Deployment Planner tool for Hyper-V and VMware environments. The tool collects disk write data for a few days and reports the required bandwidth, target disk type and storage account needs. The measurement should cover the busiest business periods (month-end close, nightly batch jobs). Measuring a quiet week is misleading.
Replication policy
The policy is defined in the vault and assigned to protected servers. The values set:
- Copy frequency (Hyper-V): 30 seconds (except premium storage) or 5 minutes. For VMware and physical servers, replication is continuous; crash-consistent points are created on the order of minutes.
- Recovery point retention: Determines how far back you can go during failover. Long retention means snapshot storage cost for each point.
- Application-consistent snapshot frequency: Taken with VSS (Windows) or pre/post scripts (Linux). On database servers, an application-consistent point every few hours shortens database recovery time after failover.
Instead of assigning a single policy to all servers, define at least two policies: one with short intervals and frequent application-consistent points for database and application servers, and a sparser one for file and auxiliary servers. For which RPO/RTO tier ASR fits into, see the choosing technology by RPO/RTO target article.
Network design: addressing, DNS and connectivity
The most time-consuming work in ASR projects isn’t replication, it’s the network. After failover, servers come up in Azure; users, branches and other servers need to reach them. A server being powered on doesn’t mean the service is up.
IP addressing
There are two approaches. Keeping IPs: A virtual network with the same address range as the on-premises subnet is prepared in Azure, and servers come up with the same IP. Hard-coded IP dependencies in applications stop being a problem; but the same range can’t be routed over VPN while it is active both on premises and in Azure, so during failover the whole subnet must be moved and routing changed. Changing IPs: A different address range is used in Azure; the target IP of each replicated server is defined in the “Compute and Network” settings, and a DNS update is required.
DNS and Active Directory
For failed-over servers to authenticate, a domain controller/DNS server is needed in Azure. The path we recommend is to keep a DC in Azure that runs continuously and stays current through AD replication; a DC replicated with ASR should be used only in an isolated test failover network. The Azure virtual network’s DNS setting is configured to point to this DC. Lowering the TTL values of critical records in advance lets clients learn the new address quickly in the IP change scenario. Leaving the virtual network’s DNS setting at the Azure default is one of the first things to blow up in a drill.
Connectivity
For users to access servers in Azure, a site-to-site VPN or ExpressRoute must be set up in advance. Creating a VPN gateway in Azure alone can take 30–45 minutes or longer; leaving this until the disaster directly breaks the RTO. If head office is unreachable, backup tunnels through which branches can connect directly to Azure should also be planned. Replication traffic goes to Azure’s public endpoints by default; if traffic should flow over a private network, a private endpoint configuration can be used.
Recovery plans
A recovery plan is the object that divides servers into groups and sets the failover order. Groups are powered on in sequence; manual steps or Azure Automation runbooks can be added before and after each group.
A typical plan consists of these groups:
- Group 1 – Infrastructure: Dependencies such as additional DNS and file servers. Script afterwards: update DNS records.
- Group 2 – Database: SQL Server servers. Manual step afterwards: verify that the databases are ONLINE.
- Group 3 – Application: ERP and application servers. Script afterwards: check the state of the services.
- Group 4 – Front end: Web and RDS servers. Script afterwards: public IP or load balancer assignment.
Scripts make the plan repeatable. But every script also runs during test failover; write steps that change production records (for example a public DNS update) conditionally based on the failover type. Otherwise a drill points production DNS to the test environment. How the technical team will run the plan and who approves which step is documented in the disaster recovery runbook. Put the conditions for activating replication, the people responsible and the order of return in writing within a disaster recovery plan.
Drills with test failover
Test failover creates virtual machines in Azure from the chosen recovery point without interrupting replication. For the drill to be worth anything, the test network must be completely isolated from production: a separate virtual network, with no VPN or peering connection. Otherwise two servers with the same name and IP conflict.
# Test failover through a recovery plan with Az PowerShell
$vault = Get-AzRecoveryServicesVault -Name "rsv-dr-weu" -ResourceGroupName "rg-dr"
Set-AzRecoveryServicesAsrVaultContext -Vault $vault
$rp = Get-AzRecoveryServicesAsrRecoveryPlan -Name "RP-ERP"
$vnet = Get-AzVirtualNetwork -Name "vnet-dr-test" -ResourceGroupName "rg-dr"
Start-AzRecoveryServicesAsrTestFailoverJob -RecoveryPlan $rp `
-Direction PrimaryToRecovery -AzureVMNetworkId $vnet.Id
# Clean up the test resources after verification
Start-AzRecoveryServicesAsrTestFailoverCleanupJob -RecoveryPlan $rp
What to measure in the drill:
- The time from the start of failover until the last group is up (actual RTO).
- The difference between the timestamp of the chosen recovery point and the time of the drill (actual RPO).
- User tests: sign-in, creating a record in ERP, reporting, printing.
- The duration of the manual steps, and which steps were missing from the document.
A real failover has extra steps: the failover is finalized with commit, replication is set up in the reverse direction from Azure to on premises with re-protect once the source is reachable again, and failback is done in a suitable maintenance window. Failback also has a data transfer and a downtime window. Put that in the plan too.
Cost items and alternatives
ASR cost isn’t a single item; since prices vary by region and over time, work out the items with the Azure Site Recovery pricing page and the Azure pricing calculator:
- ASR fee per protected server (according to the pricing page at the time of writing, the first 31 days of each server are free).
- Replica managed disks and snapshot storage for recovery points.
- Cache storage account transactions.
- Compute: VMs running only during test or real failover.
- Network: VPN/ExpressRoute gateway, data egress from Azure during failback.
- Licensing: How Windows Server and SQL Server licenses will be covered in Azure (Azure Hybrid Benefit, DR rights) should be evaluated separately.
| Criterion | Azure Site Recovery | Backup software’s Azure target |
|---|---|---|
| Purpose | Fast failover | Long-term copy |
| Typical RPO | Minutes | Last backup (hours) |
| Typical RTO | Minutes – a few hours | Hours (data is restored) |
| Rollback window | Hours – days | Weeks – years |
| Against ransomware | Limited | Strong with immutable storage |
These two approaches aren’t alternatives to each other but complements: ASR for servers that need a short RTO, the backup copy for long-term retention and protection against logical corruption. A replica carries encrypted data too. If there is a second site on premises, Hyper-V Replica and Veeam replication can also be put on the table. For subscription, network and governance design on the Azure side, see our Microsoft 365, Azure, AWS and Google Cloud services page, and for the DR architecture as a whole, our backup and disaster recovery page.
Frequently asked questions
Does Azure Site Recovery also protect virtual machines in Azure?
Yes. ASR can replicate Azure virtual machines from one Azure region to another. In this scenario no on-premises appliance or provider is installed; the Mobility service extension is added to VMs automatically, and network mapping and the recovery plan are still defined in the vault.
Does a test failover affect the production environment?
Since replication isn’t interrupted, the source servers aren’t affected. But if the test virtual network is connected to the production network through VPN or peering, servers with the same name and IP conflict; isolate the test network completely.
Can servers keep the same IP address after failover?
Yes, if a virtual network with the same address range is used in Azure. In that case the whole subnet must be failed over and routing changed accordingly; the same range can’t be actively routed on both sides at the same time.
What is the most overlooked item in ASR cost?
Usually the compute and licensing cost after failover, and data egress during failback, are forgotten. In normal times, the amount paid consists of replica disks and the protected server fee; in a real disaster, all VMs start running in Azure.