Migrating VMware Workloads to Google Cloud

Migrating VMware Workloads to Google Cloud

Reading time1 min
#Cloud#Migration#VMware#GCP#GoogleCloud

Migrating VMware Workloads to Google Cloud

There are two different destinations for vSphere VMs in Google Cloud, and choosing between them is the first decision:

  • Google Cloud VMware Engine (GCVE): a dedicated VMware Cloud Foundation private cloud (vSphere, vSAN, NSX) running on Google hardware. VMs move without conversion and keep their IP addresses.
  • Compute Engine: VMs are converted into native Google Cloud instances with Migrate to Virtual Machines. No VMware layer remains.

GCVE is the fast exit from a data center. Compute Engine is the long-term cloud-native target. Many migrations do both: lift everything to GCVE first, then convert or rebuild workloads one by one.

Assess first

Collect an inventory before choosing. Google Cloud Migration Center imports RVTools exports (the vInfo and vCPU tabs) or collects data with its discovery client through the vSphere API. It gives you VM counts, sizes, utilization and Compute Engine sizing and cost estimates.

For each VM, note:

  • Guest OS and version. Migrate to Virtual Machines supports a list of Windows and Linux versions; anything outside it is a candidate for GCVE or a rebuild.
  • Dependencies: which VMs talk to which, plus hard-coded IP addresses, licensing servers and anything bound to a MAC address.
  • Licensing. Windows Server can use Google's pay-as-you-go licenses on Compute Engine. Bringing your own Windows or SQL Server licenses usually requires sole-tenant nodes.
  • Disks that are huge or change very fast: they dominate replication time.

Option A: Google Cloud VMware Engine

You provision a private cloud in a region, connect it to your VPC and your data center, and migrate with VMware HCX, which is part of the service.

HCX offers several migration types: bulk migration (parallel, with a reboot at switchover), vMotion (live, one VM at a time) and replication-assisted vMotion (parallel replication, live switchover). With HCX network extension, the on-premises layer 2 segments stretch into GCVE, so VMs keep their IPs and you can move them in waves without readdressing. Remove the extension once a segment is fully migrated, so routing goes directly through Google Cloud.

Licensing changed after Broadcom acquired VMware. Since late 2025, new VMware Engine capacity is sold as nodes where you bring your own VMware Cloud Foundation subscription purchased from Broadcom. Check the current terms before you plan around GCVE.

Option B: Compute Engine with Migrate to Virtual Machines

Migrate to Virtual Machines replicates vSphere VMs into Google Cloud and converts them during cut-over.

  1. Deploy the Migrate Connector. It is an OVA appliance you deploy in vCenter. It reads VM disks through vSphere APIs and sends data to Google Cloud over HTTPS (port 443), over the internet, Cloud VPN or Cloud Interconnect.
  2. Register the source in Migrate to Virtual Machines and group VMs into migration waves.
  3. Start replication. The first sync copies full disks, then incremental syncs follow on a schedule while the VM keeps running on-premises.
  4. Configure target details: project, zone, machine type, network, subnet, internal IP, network tags, disk type, license type.
  5. Test clone. Start a copy of the VM in a test VPC without touching the source. Check boot, application, licensing and monitoring agents.
  6. Cut-over. The connector shuts down the source VM, runs a final sync and creates the Compute Engine instance. During conversion it adapts the guest OS for Google Cloud: drivers, the guest environment, network configuration.
  7. Finalize after validation, which cleans up the replication data.

Downtime per VM is roughly the final sync plus boot time, usually minutes. For single disk images (VMDK, VHD) use image import, and for OVA/OVF appliances use machine image import; both are part of Migrate to Virtual Machines. The older gcloud compute images import tool reached end of life in July 2025.

Networking and connectivity

Set up the target environment before the first wave:

  • Hybrid connectivity: HA VPN for smaller migrations, Dedicated or Partner Interconnect for large replication volumes and stable latency.
  • IP plan: on Compute Engine VMs get IPs from VPC subnets. Either reserve the old addresses in matching subnet ranges or plan readdressing and DNS updates per wave.
  • Firewall: NSX or perimeter firewall rules translate into VPC firewall rules or network firewall policies, scoped by network tags or service accounts.
  • DNS: lower TTLs ahead of each wave, and make on-premises and cloud DNS resolve each other (Cloud DNS forwarding zones and inbound server policies).

Data and databases

Large databases often migrate better at the application level than as VM disks. Database Migration Service replicates MySQL, PostgreSQL and SQL Server into Cloud SQL, and native replication (log shipping, Always On availability groups, MySQL replication) works into Compute Engine VMs. File shares can move to Filestore with Storage Transfer Service agents, or stay on a VM. For tens of terabytes over a thin link, Transfer Appliance avoids weeks of network transfer.

Run it in waves

  1. Pilot: two or three low-risk VMs, end to end, including backup and monitoring.
  2. Group the rest by dependency so that chatty VMs move together.
  3. For each wave: replicate, test clone, announce the window, cut over, validate, keep the source powered off but intact for the rollback period.
  4. After the rollback period, delete the source VMs and release the on-premises capacity.

After the move

  • Rightsize with the recommendations Compute Engine provides after a few weeks of real utilization.
  • Replace VMware-era tooling: snapshot schedules or Backup and DR instead of VM-level backup products, Ops Agent instead of in-guest monitoring tied to vCenter.
  • Remove VMware Tools from converted Linux and Windows guests if it is still installed.
  • Consider managed services for the components that were VMs only because they had to be: Cloud SQL, Memorystore, GKE.

Common pitfalls

  • Starting replication without enough bandwidth, then discovering the initial sync takes weeks.
  • Forgetting dependencies on hard-coded IPs, local license servers or Active Directory sites.
  • Unsupported guest OS versions found on cut-over day instead of during assessment.
  • No rollback plan: keep the source VM until the target has run through at least one business cycle.

Summary

  • Lift as-is with VMware intact: GCVE plus HCX, IPs preserved via network extension.
  • Convert to native instances: Migrate to Virtual Machines with the Migrate Connector, replicate, test clone, cut over.
  • Assess with Migration Center (RVTools import or discovery client).
  • Hybrid connectivity, IP plan and firewall rules before wave one.
  • Pilot, then dependency-based waves, each with a rollback window.
    Migrating VMware Workloads to Google Cloud