The best cross-geography disaster recovery (DR) solution depends on the layer you need to replicate. For VMs and servers, HPE Zerto, Veeam, AWS Elastic Disaster Recovery, and Azure Site Recovery lead. For databases, native global options such as Amazon Aurora Global Database, Azure Cosmos DB, Google Cloud Spanner, DynamoDB global tables, and CockroachDB give the lowest data loss. For object storage, turn on each cloud’s cross-region replication and pair it with versioning or object lock. Most teams combine one tool per layer with immutable backups and a traffic-failover layer, and platforms that run the application active-active across regions and clouds, such as Control Plane, remove the separate failover step for compute while still needing a replicated database.
Disclosure: Control Plane publishes this guide.
TL;DR
Each layer of an application needs its own replication method, so a cross-geography DR plan is a set of tools.
- Pick one tool per replication layer and combine them into one tested plan.
- For VM-level replication, HPE Zerto and Veeam reach across data centers, clouds, and on premises sites. AWS Elastic Disaster Recovery and Azure Site Recovery are strongest inside their own clouds.
- For the lowest RPO on databases, use database-native global replication such as Aurora Global Database, Cosmos DB, Spanner, DynamoDB global tables, or CockroachDB.
- For object storage, enable cross-region replication on Amazon S3, Azure Storage, or Google Cloud Storage, with versioning turned on.
- Replication copies corruption too. Pair every replication tool with immutable backups from Rubrik, Commvault, Veeam, or your cloud’s backup vault.
- Traffic failover is a separate layer. DNS or a global load balancer has to move users to the healthy region.
- Test failover on a schedule.
- Control Plane runs workloads active-active across AWS, GCP, Azure, and Oracle Cloud Infrastructure (OCI), which removes the compute failover step. It does not replicate volume data between locations or replace a backup tool.
What are the best disaster recovery solutions for cross geography replication?
The leading solutions split by layer: VM-level tools replicate servers and block storage, database engines replicate transactions, object storage services replicate files, Kubernetes tools back up cluster state, and backup platforms keep immutable copies.
| Solution | What it replicates | Replication mode | Published RPO / RTO | Reach | Main limit |
|---|---|---|---|---|---|
| HPE Zerto | VMs and applications as consistency groups | Continuous, journal-based | RPO seconds; RTO minutes (Zerto) | VMware vSphere and cloud targets including AWS and Azure | Replicates at the hypervisor level; databases still need their own replication for zero data loss |
| Veeam Data Platform | VMs, physical servers, NAS, cloud workloads | CDP into vSphere; snapshot replication elsewhere | RPO seconds with CDP; minutes with snapshot replication (Veeam) | On premises, hybrid, and cloud | CDP targets must be VMware vSphere |
| AWS Elastic Disaster Recovery | Servers at the block level | Continuous block replication to a staging area | RPO seconds; RTO minutes (AWS) | Sources on premises, in other clouds, or in other AWS Regions; recovery in AWS | Recovery always happens in an AWS Region |
| Azure Site Recovery | Azure VMs, VMware and Hyper-V VMs, physical servers | Continuous for Azure and VMware VMs; as often as every 30 seconds for Hyper-V | Crash-consistent recovery points every 5 minutes (Microsoft) | Any other Azure region; on premises into Azure | Cloud target must be Azure |
| Google Cloud Backup and DR | Google Cloud VMs, databases, file systems | Scheduled or on-demand backups to vaults | RPO set by the backup schedule | Google Cloud regions | Backup-based; recovery needs a restore step |
| Rubrik | VMs, databases, SaaS apps, cloud workloads | Incremental backup with immutable snapshots | RPO set by backup frequency | On premises, cloud, and SaaS | Snapshot backup; recovery needs a restore step |
| Commvault Cloud | VMs, databases, files, containers, SaaS apps | Backup with a replication option | RPO set by schedule or replication frequency | On premises, hybrid, and multi-cloud | Policy setup across environments takes planning |
| Database-native global replication (Aurora Global Database, DynamoDB global tables, Cosmos DB, Spanner, CockroachDB) | Database rows and transactions | Synchronous (Spanner, CockroachDB) or asynchronous (Aurora Global Database; DynamoDB and Cosmos DB by default, each with a strong option) | RPO zero with synchronous options; seconds with asynchronous ones | Same cloud, except CockroachDB, which runs across clouds | Each engine has its own consistency, DDL, and feature limits |
| Object storage replication (S3 Cross-Region Replication, Azure GRS/RA-GRS/GZRS, Google Cloud Storage dual-region and multi-region) | Objects and metadata | Asynchronous | Minutes to hours; 15 minutes with S3 Replication Time Control or Google Cloud Storage turbo replication | Same cloud, another region | Does not cover application state, configuration, or search indexes |
| Velero and Veeam Kasten | Kubernetes resources and persistent volumes | Scheduled backups | RPO set by the schedule | Another cluster, region, or cloud, with the right data mover | Volume snapshots do not move across providers on their own |
| Control Plane | Running workloads (serverless, standard, stateful, cron, VM), active-active across locations | Active-active; DNS routes to healthy locations | Not published | AWS, GCP, Azure, Oracle Cloud, your own Kubernetes clusters and servers, on premises | Does not replicate volume data between locations; not a backup product |
HPE Zerto
HPE Zerto provides continuous data protection through journal-based replication at the hypervisor level. It captures every write and streams it to a target site, building a journal you can recover from at any point in time inside the journal window. Zerto states RPOs of seconds and RTOs of minutes.
Zerto replicates groups of VMs as one consistency group, which keeps multi-tier applications consistent with each other. It supports VMware vSphere and cloud targets including AWS and Azure, so it suits estates that mix data centers and clouds.
Because Zerto works at the hypervisor level, it does not replicate databases at the transaction level. Databases that cannot lose any committed data still need a database-native replication layer underneath.
Veeam Data Platform
Veeam combines backup, replication, and continuous data protection (CDP) in one platform. Veeam CDP gives RPOs in seconds, and its targets must be a VMware vSphere cluster or host. Universal CDP in Veeam Backup & Replication 13 can replicate physical, virtual, and cloud machines into vSphere. Other targets use snapshot-based replication with RPOs in minutes.
Veeam’s orchestration automates failover sequences and runs test failovers that do not touch production. It supports targets across AWS, Azure, Google Cloud, and on premises infrastructure, which gives it broad multi-cloud failover coverage. Veeam also offers immutable backup repositories for ransomware recovery.
AWS Elastic Disaster Recovery
AWS Elastic Disaster Recovery (AWS DRS) continuously replicates source servers at the block level into a low-cost staging area in an AWS Region. Sources can be on premises servers, VMs in other clouds, or EC2 instances in another Region. When you fail over, AWS DRS launches recovery instances from the replicated data. AWS states that DRS enables RPOs of seconds and RTOs of minutes.
DRS supports point-in-time recovery, so you can recover to a state from before corruption began. After the incident, it can replicate data back and fail back to the original site.
Recovery always happens in an AWS Region. AWS DRS can fail back to an on premises source, but it cannot recover into Azure or Google Cloud, so a multi-cloud DR plan needs another tool for non-AWS targets.
Azure Site Recovery
Azure Site Recovery (ASR) replicates Azure VMs to any other Azure region and replicates on premises VMware VMs, Hyper-V VMs, and physical servers into Azure. It replicates Azure VMs and VMware VMs continuously and Hyper-V VMs as often as every 30 seconds. It creates crash-consistent recovery points every five minutes.
Recovery plans set the start order of VMs, run scripts, and integrate with Azure Automation. ASR pairs with Azure Front Door or Azure Traffic Manager for traffic failover, and test failovers are built in.
The cloud target must be Azure. Check the Azure Site Recovery support matrix for your operating systems and workloads first. Cloud disaster recovery solutions are compared in more depth in a separate roundup.
Google Cloud Backup and DR Service
Google Cloud Backup and DR Service protects Google Cloud VMs, databases, and file systems with backup vaults that hold scheduled or on-demand backups. Vaults can sit in a different region from the source, which supports cross-geography recovery.
The service is backup-based. RPO follows the backup schedule, and RTO depends on how long the restore takes. It fits Google Cloud environments that accept backup-level RPOs and want a managed service.
Rubrik
Rubrik delivers backup with immutable, air-gapped snapshots that ransomware cannot modify or delete. It protects VMs, databases, SaaS applications, and cloud workloads, with copies on Rubrik clusters and in cloud archives.
Rubrik is a backup product, so recovery requires a restore step and RTOs are longer than with journal-based or CDP tools. Its strength is clean recovery after ransomware and long-term retention. Pairing Rubrik’s immutable copies with a continuous replication tool gives you fast failover and a clean recovery point if the replicated data is corrupted.
Commvault Cloud
Commvault Cloud protects VMs, databases, files, containers, and SaaS applications across on premises, hybrid, and multi-cloud environments. It offers backup and replication with policy-driven scheduling, retention, and cross-region copies.
Setting up replication policies across several environments takes careful planning. Like Rubrik, Commvault’s main value is backup and retention, and it pairs well with a continuous replication tool for workloads that need low RPOs.
Database-native global replication
Database-native global replication usually gives a lower RPO than VM-level or storage-level tools because it replicates committed transactions.
- Amazon Aurora Global Database replicates an Aurora MySQL or PostgreSQL cluster from one primary Region to up to 10 secondary Regions, with replication lag typically under one second. An unplanned failover typically promotes a secondary within a few minutes, with RPO measured in seconds. A planned switchover loses no data.
- DynamoDB global tables replicate items across AWS Regions with active-active writes. The default mode is multi-Region eventual consistency. A multi-Region strong consistency mode gives zero RPO but does not support transactions or TTL.
- Azure Cosmos DB offers five consistency levels. Strong consistency gives zero RPO but requires a single write region. Accounts with multi-region writes use weaker levels, with an RPO under 15 minutes for session, consistent prefix, or eventual consistency.
- Google Cloud Spanner replicates synchronously across regions with strong consistency, so committed data survives a region failure. Write latency grows with the distance between regions.
- CockroachDB runs on any cloud. With the REGION survival goal and at least three regions, it replicates synchronously across regions and clouds and loses no committed data when one region fails. Write latency depends on the round trip to a majority of replicas.
Aurora Global Database and DynamoDB are AWS-only, and Cosmos DB and Spanner each stay inside their own cloud. CockroachDB runs anywhere and takes more operational work. DDL handling, conflict resolution, and consistency guarantees differ by engine. Failing over a database between cloud regions covers the runbook step by step.
Object storage replication
Every major cloud replicates object storage across regions asynchronously.
- Amazon S3 Cross-Region Replication (CRR) copies objects and metadata to a bucket in another AWS Region. S3 Replication Time Control adds an SLA to replicate 99.99% of new objects within 15 minutes.
- Azure geo-redundant storage (GRS, RA-GRS, GZRS) copies data to the paired region. RA-GRS adds read access in the secondary region before a failover.
- Google Cloud Storage dual-region and multi-region buckets replicate objects to a second region asynchronously. Default replication targets 99.9% of new objects within one hour and 100% within 12 hours. Turbo replication, available for dual-region buckets only, targets a 15-minute RPO.
Object storage replication covers files, images, logs, and other unstructured data. Application configuration, search indexes, and in-memory state need their own plan.
Velero and Veeam Kasten for Kubernetes
Kubernetes workloads need their own DR tooling because VM-level replication does not capture cluster resources such as Deployments, ConfigMaps, Secrets, and custom resources.
- Velero backs up Kubernetes resources and persistent volume snapshots to object storage and can restore into another cluster, region, or cloud. Volume snapshots do not move across providers or across AWS and Azure regions, so cross-region and cross-cloud restores need Velero’s file system backup or snapshot data mover.
- Veeam Kasten (formerly Kasten K10) provides application-aware backup and mobility for Kubernetes, with policy-driven schedules, encryption, and immutable backups. It can move applications across clusters and cloud providers when you export snapshot data.
Both tools work from scheduled backups, so RPO follows the schedule and RTO follows restore time. For lower RPOs, keep persistent data in a replicated database and use Velero or Kasten for cluster resources.
Control Plane
Control Plane is the unbreakable platform for running production workloads. It runs serverless, standard, stateful, and cron workloads natively on AWS, GCP, Azure, and Oracle Cloud Infrastructure (OCI), VM workloads on locations joined through Control Plane’s Managed Kubernetes, and workloads on Kubernetes clusters you already run or your own machines, including on premises, active-active across regions and clouds.
A Global Virtual Cloud (GVC), Control Plane’s named set of locations, can span Control Plane-managed regions on AWS, GCP, Azure, and Oracle Cloud, plus your own clusters and servers. A workload runs in every location of its GVC, active-active, so there is no standby compute to start during a regional outage. Control Plane gives each workload a canonical endpoint, its default address. That endpoint uses DNS to send users to the nearest location where the workload passes its readiness probes, so traffic moves away from an unhealthy location without a manual DNS change.
Control Plane does not replicate volume data between locations. Each location’s volume set, Control Plane’s persistent storage for stateful workloads, holds its own data. Cross-location data protection comes from a replicated database or from backups. Control Plane’s template catalog includes multi-location database templates that run one cluster across the locations of a GVC: pgEdge (active-active PostgreSQL with multi-master logical replication; DDL does not replicate and every table needs a primary key) and a MongoDB replica set that can survive losing a location when spread over three locations.
Control Plane is not a backup product and does not provide immutable backup vaults, so pair it with a backup tool for ransomware recovery. Control Plane handles Day 2 operations for you, including autoscaling, built-in observability, and security. Control Plane is PCI DSS Level 1, SOC 2 Type II, HIPAA, and GDPR compliant.
How do you choose a cross-geography DR solution?
Start from what you are replicating, how much data loss you can accept, and how fast you need to recover.
| What you are replicating | Best fit | Why | Watch out for |
|---|---|---|---|
| VMware or Hyper-V VMs | HPE Zerto, Veeam, or Azure Site Recovery | Continuous or journal-based replication with orchestrated failover | Veeam CDP targets must be vSphere; ASR recovers into Azure |
| Cloud VMs in one hyperscaler | AWS Elastic Disaster Recovery (AWS) or Azure Site Recovery (Azure) | Native integration and managed failover orchestration | Recovery stays in that cloud; Google Cloud Backup and DR is backup-based with a restore step |
| Relational database | Aurora Global Database, Spanner, CockroachDB, or PostgreSQL with pgEdge | Transaction-level replication with the lowest RPO; synchronous options exist | Synchronous replication adds write latency that grows with distance |
| NoSQL database | DynamoDB global tables or Cosmos DB | Active-active writes across regions | Cosmos DB multi-region writes cannot use strong consistency; DynamoDB strong mode drops transactions and TTL |
| Object storage | S3 Cross-Region Replication, Azure GRS or GZRS, Google Cloud Storage dual-region or multi-region | Built in, little to operate | Asynchronous by default; turn on versioning so deletes and overwrites can be undone |
| Kubernetes workloads and volumes | Velero or Veeam Kasten | Backs up cluster resources and volumes and restores to another cluster | Backup-based RPO; cross-cloud restores need file-level backup or exported snapshot data |
| A whole application that must keep serving during a regional outage | Active-active compute plus a replicated database plus immutable backups | No failover step for compute; DNS routes around the failed region | Database replication, backups, identity, and secrets still need their own plan |
| Data with residency rules in one country | Any tool that lets you choose target regions explicitly | Keeps copies inside the approved jurisdiction | Check that every target region, including backup vaults, sits in that country |
RPO and RTO targets
RPO is the maximum acceptable data loss, measured as time before the failure. RTO is the maximum acceptable downtime. A workload with a 1-second RPO and a 5-minute RTO needs continuous replication and automated failover. A workload with a 24-hour RPO and a 4-hour RTO can often use nightly backups and a manual restore, if the restore finishes within 4 hours.
Synchronous vs. asynchronous replication
Synchronous replication acknowledges a write only after a remote site or a quorum of replicas confirms it, so each write waits at least one network round trip. Round-trip times between continents are roughly 80 to 300 milliseconds on Azure’s network, according to Microsoft’s published latency statistics, and some protocols need more than one round trip per write.
Asynchronous replication commits locally and streams changes to the other site in the background. It adds no write latency, but data written during the replication lag can be lost in a failure.
DR patterns
Four common patterns trade cost against recovery speed.
- Backup and restore keeps backups in a remote region. It costs the least and recovers the slowest.
- Pilot light keeps core pieces, such as database replicas and DNS records, running in the target region, and starts compute only during failover.
- Warm standby runs a scaled-down copy of the whole environment in the target region and scales it up during failover.
- Active-active runs the full application in several regions at once and serves users from the nearest healthy one. It has the lowest RTO. RPO depends on how the database replicates.
What does cross-geography replication not protect?
Replication alone leaves eight risks open, each with its own fix.
| Risk | Why replication alone fails | What to add |
|---|---|---|
| Ransomware or corruption | Replication copies corrupted data within seconds, and journal-based rollback only reaches back as far as the journal window | Immutable backups with long enough retention (Rubrik, Commvault, Veeam immutable repositories, cloud backup vaults) |
| Accidental deletes | Most replication tools copy deletes along with writes | Point-in-time recovery, soft delete, object versioning, or versioned backups |
| DNS and traffic still pointing at the failed region | Replication moves data and leaves routing unchanged | DNS failover, a global load balancer, or active-active routing |
| Identity and secrets missing in the target region | IAM roles, certificates, API keys, and secrets may not exist in the secondary region | Pre-provision identity configuration and secrets in every target region |
| Capacity not available in the target region during a regional event | Many organizations fail over to the same nearby regions at once | Reserve capacity, use more than one target region, or run active-active |
| Data residency violation in the target geography | A target region in another country may break transfer rules or contractual residency commitments | Pin replicas and backups to approved regions |
| Untested runbooks | Untested runbooks often miss steps, permissions, or dependencies | Test failover regularly without taking production down |
| Inter-region and egress transfer costs | Replication generates continuous cross-region traffic | Budget for transfer charges, compress replication streams, and check each cloud’s pricing |
How do you keep replicated data in the right country?
GDPR, PIPEDA, and Australia’s Privacy Act do not ban storing personal data abroad. GDPR limits transfers outside the EEA to adequate countries or approved safeguards, and PIPEDA and Australian Privacy Principle 8 keep the sending organization accountable for how the data is protected overseas. Some sector rules, contracts, and customer commitments do require in-country storage, and in those cases replicating to a region in another country can break the rule.
Azure pairs most regions with another region in the same geography (for example, Canada Central and Canada East), though some pairs cross borders and many newer regions have no pair. AWS lets customers choose target Regions explicitly. Google Cloud Storage lets you build a dual-region from two regions in the same country, such as Montréal and Toronto.
Keep encryption keys in a key management service in the same jurisdiction as the data, so the data cannot be decrypted with keys held elsewhere. Keeping customer data in country during multi-region failover walks through the setup.
How Control Plane fits
Control Plane runs workloads active-active across regions and clouds, which removes the separate failover step for the compute layer.
In a standby design, you replicate VMs or containers to a second region, maintain DNS failover rules, and depend on capacity being available in the standby region during the event. On Control Plane, the standby region is replaced by more active locations in the same GVC, and traffic already flows to every healthy location.
This architecture still needs a replicated database, such as the pgEdge and MongoDB templates or a database-native option. Backups come from a separate tool. Identity, secrets, and failover testing stay on the team’s plan.
FAQ
These answers cover RPO, cost, ransomware, databases, Kubernetes, and traffic failover.
What are the best disaster recovery solutions for cross geography replication?
It depends on the layer. Use VM replication tools such as HPE Zerto, Veeam, AWS Elastic Disaster Recovery, or Azure Site Recovery for servers, database-native replication for databases, and cloud object replication for files, plus immutable backups and traffic failover. Platforms such as Control Plane run compute active-active, so the remaining work is a replicated database and backups.
What is the difference between backup and replication?
Backup keeps periodic copies for retention and recovery from corruption, deletion, or ransomware. Replication keeps a near real-time copy in another location for fast failover during an outage. Replication copies corruption along with good data, so most DR strategies need both.
Is synchronous replication possible across regions?
Yes, with a latency cost. Google Cloud Spanner and CockroachDB replicate synchronously across regions. Microsoft’s published Azure latency data shows roughly 80 milliseconds between the US East Coast and Western Europe and over 200 milliseconds between the US East Coast and Australia, and every synchronous write waits at least one such round trip. Applications that cannot absorb that latency use asynchronous replication and accept a small RPO window.
What RPO and RTO can cross-region replication achieve?
Cross-region replication can reach zero RPO with synchronous databases and an RPO of seconds with continuous replication tools such as HPE Zerto and AWS Elastic Disaster Recovery, which also state RTOs of minutes. Backup-based tools reach RPOs of minutes to hours depending on the schedule. Active-active architectures skip compute failover, so recovery time comes down to health-check intervals and DNS TTLs.
Does replication protect against ransomware?
Not on its own. Replication copies encrypted or corrupted files to the other site within seconds. Journal-based tools can roll back a short window, but attacks often go undetected for longer. Keep immutable backups long enough that a clean copy from before the infection still exists.
How much does cross-region replication cost?
Cross-region replication cost is the sum of inter-region data transfer, extra storage, standby compute, and tool licenses. Active-active designs pay for compute in every region and avoid paying for idle standby capacity. Check each cloud’s transfer pricing and each vendor’s licensing before you size the plan.
How do I replicate a database across clouds?
Use a database that runs and replicates on every cloud, such as CockroachDB or PostgreSQL with pgEdge, or use change data capture between managed databases. On Control Plane, the pgEdge and MongoDB templates run one cluster across the locations of a GVC.
How do I replicate Kubernetes workloads across regions?
Back up cluster resources and volumes with Velero or Veeam Kasten, then restore them into a cluster in the target region. Moving volume data between regions or clouds takes a file-level backup or exported snapshot data. Databases that need second-level RPOs should replicate natively instead.
How does Control Plane handle cross-region disaster recovery?
Control Plane runs every workload in every location of its Global Virtual Cloud, active-active, across AWS, GCP, Azure, Oracle Cloud, and your own clusters or servers. Each workload’s endpoint routes users away from locations that fail readiness checks. Volume data stays in each location, so teams pair Control Plane with a replicated database and an immutable backup tool.
Do I need a separate traffic failover layer?
Yes, for anything your platform does not route. Most replication tools leave traffic routing to a separate DNS or load-balancing layer, such as Amazon Route 53, Azure Traffic Manager, Azure Front Door, or AWS Global Accelerator. On Control Plane, each workload’s canonical endpoint uses DNS to route users away from unhealthy locations, so workloads that run only on Control Plane do not need a separate DNS failover rule. Services outside Control Plane still need one.
Ready to run your workloads active-active across regions and clouds? Talk to the Control Plane team or sign up free.
Originally published on the Control Plane blog.
Top comments (0)