EXAMTOPICS / TECHNOLOGY INSIGHTS

Converting VMware RDMs to VMDKs: Steps and Risks

Converting a VMware raw device mapping (RDM) into a virtual machine disk (VMDK) changes how a guest accesses storage, not the application data stored on that disk. The reason for migrating might be to standardize backups, simplify storage management, retire a physical storage array, or remove a shared-LUN dependency. The conversion method must follow the RDM compatibility mode, application requirements, and available maintenance window. A VMDK conversion is not a substitute for a database-consistent backup or a tested rollback plan.

An RDM gives a virtual machine access to a storage LUN through a mapping file. The distinction between virtual compatibility mode (RDM-V) and physical compatibility mode (RDM-P) is decisive: their snapshot and migration capabilities differ, as do the risks of changing a disk while its guest application is operating. Before using a wizard, identify exactly which mode and storage object the machine has. The wider raw device mapping comparison explains why the two modes exist and where their operational limitations matter.

Confirm the disk mode and application constraints

Start by recording the virtual machine, datastore, RDM mapping file, physical LUN identifier, SCSI controller, unit number, disk size, provisioning type, guest volume label, file system, and application owner. Use vCenter and the guest operating system together. A disk called “Hard disk 3” in vCenter is not enough evidence to identify the correct guest volume, especially when a VM contains several disks of the same capacity.

For an RDM-V, Storage vMotion can convert the disk into a VMDK by selecting an appropriate virtual disk format in the advanced migration options. Broadcom documents that a virtual-mode RDM can be converted while the VM is powered on, subject to normal compatibility and placement constraints. That permission does not prove the workload can tolerate the move. Storage arrays, application clusters, backup software, guest I/O patterns, and maintenance policy may impose tighter limits.

For an RDM-P, a conversion generally requires downtime for the affected VM or at least for the RDM backing LUN. Treat this as a planned storage change with coordinated application quiescence, rather than telling operators to repeat the live RDM-V steps. Physical mode is commonly found in more specialized configurations. Shared-disk clustering and vendor-supported integrations deserve their own compatibility assessment; replacing the disk type without checking those dependencies can break a supported design.

Broadcom’s RDM-to-VMDK conversion guidance is a primary reference for the virtual-versus-physical procedure. Use the vendor’s current product documentation as the procedural authority for your exact vSphere version and target environment, not a copied series of clicks from an unrelated release.

Build a change plan that distinguishes data copies from cutover

The plan should list the current mapped LUN, target datastore, free capacity, expected I/O, migration method, change window, dependency owners, and stop conditions. A block-by-block conversion preserves guest disk structures, so whether the guest uses NTFS, XFS, ext4, or another supported file system is not by itself the deciding factor. The guest may still require a volume-rescan or validation after the virtual disk is attached. In contrast, application consistency, access conflicts, and storage-path health are genuinely material.

Reserve enough usable datastore space for the resulting VMDK and operational headroom; do not plan against a nominally free capacity figure that ignores thin-provisioning overcommit, snapshots, replication, and existing growth. Confirm that all involved hosts can reach the necessary datastores, that the migration account has the required privileges, and that the storage policy and encryption requirements will still be met afterward. If target storage has materially different performance or failure domains, measure those differences before cutover.

Take an application-aware backup that can be restored independently of the RDM. Check its completion status and recovery scope, including database logs or other consistency dependencies. If the VM is clustered, determine which nodes can access the disk and who will own it during the conversion. Arrange a rollback that does not depend on deleting the only valid copy of production data.

Convert virtual-mode RDMs with the supported migration workflow

For a suitable RDM-V, use the supported vSphere migration workflow and inspect the per-disk advanced placement options. Explicitly select the RDM for conversion, choose a datastore with verified space and connectivity, and select the virtual disk format required for the destination. Do not assume that selecting a new datastore without changing the format necessarily converts every mapped disk. Read the migration summary carefully so the planned disk transformation is visible before execution.

Monitor vCenter tasks, datastore latency, guest I/O, storage-path errors, and application health throughout the copy. A progress bar moving forward is not evidence that latency-sensitive services remain healthy; compare the guest’s application metrics and transaction behavior with a pre-migration baseline. If the workload enters an unacceptable state or a support constraint appears, stop according to the agreed change procedure rather than repeatedly trying format combinations.

After completion, confirm the disk shows as a VMDK on the intended datastore, with the expected capacity, controller placement, and storage policy. Verify that the old physical RDM backing LUN is no longer referenced by the converted VM. Do not reclaim the LUN until every dependent guest, cluster node, backup product, and storage team agrees it is no longer required. VMFS-backed virtual disks remain subject to datastore capacity, snapshots, and array health; conversion does not remove those operational dependencies.

Plan physical-mode and clustered-disk conversions separately

A physical-mode RDM should be handled as a distinct change path. Determine whether the supported process for the environment is a cold migration, a controlled detach-and-copy sequence, or another Broadcom-documented procedure. Stop writers to the affected disk, preserve cluster ownership rules, and confirm that the source cannot receive new application writes during the copy. If the guest application uses a database or a multi-disk consistency group, coordinate it as a single recoverable unit.

For Microsoft failover clusters, Oracle clusters, or storage-vendor-specific tools, consult the supported deployment and disk-sharing matrix before converting or replacing shared storage. A seemingly successful copy can still leave the cluster unable to arbitrate ownership, discover a reservation, or meet guest-level SCSI expectations. If the target VMDK cannot support the required clustered-disk semantics on the installed vSphere release, retain the supported storage design or choose a supported alternative rather than forcing a conversion.

Record both the logical and physical disk identifiers before and after the change. Match them against the planned inventory, not against similar-looking LUN sizes. Preserve configuration exports, cluster resource configuration, and a recovery decision tree in case the target disk attaches to the wrong controller, uses a different unit number, or appears with an unexpected guest device name.

Validate application behavior before declaring success

Post-conversion validation should start at the storage layer and end at the application. Confirm that vCenter reports the expected disk backing and no active host storage errors; then verify guest volume discovery, mount points, file-system health, read/write access, and sufficient free space. Check operating-system logs for disk resets, path failures, or file-system errors. For databases, validate transactions and recovery status rather than treating a successful login screen as proof of consistency.

Compare representative workload latency, throughput, and I/O queueing against the baseline under similar load. A storage copy may have succeeded while the new datastore introduces a performance bottleneck. Review backup policy: verify that the product detects the new virtual disk and, where applicable, that a fresh backup and test restore work under the new configuration. Revisit snapshot and replication policies since an RDM and a VMDK may interact differently with those tools.

Finally, complete the change record with final datastore, disk format, storage policy, application results, any known deviations, and the actual post-change backup location. Only then approve removal of obsolete mapping files or source LUN presentation. Deleting storage on the same day as the first healthy application check eliminates recovery options at the moment they may still be needed.

Make the post-change monitoring plan time-bounded. A successful migration task only proves that one transfer completed; it does not prove there are no delayed faults. During the next production peak and the next scheduled backup, examine datastore read/write latency, guest disk queueing, database wait events, storage-array path alerts, and unexpected retry counts. Record what would justify another change window and who owns remediation. If an array is being decommissioned, make sure host zoning, replication policies, monitoring rules, and disaster-recovery scripts have all been updated to the new destination. Orphaned SAN references can turn a successful migration into a later outage when a retired LUN is finally removed.

Keep the rollback boundary explicit

Rollback should name the exact event that causes it: data inconsistency, excessive latency, cluster failure, unresolvable guest discovery, or inability to produce a recoverable backup. If the source RDM has not been modified since cutover and remains available, a properly coordinated return may be possible. If data has been written only to the target VMDK, the original RDM is stale; simply reconnecting it would lose those writes. In that case rollback depends on a planned synchronization or a tested application restore, not a storage checkbox.

A safe migration is therefore a controlled sequence of evidence: identify the disk and mode, establish application recovery, choose the supported conversion path, capture a consistent source, verify the target under load, and retire the old LUN only when recovery no longer depends on it. That sequence is more important than any single version-specific wizard label.