Storage virtualisation and datastores

What storage virtualisation means

Storage virtualisation separates the storage a workload sees from the physical disks that hold it. Disks from one or many devices are pooled, protected and carved into logical volumes. In a virtual environment this goes one step further: each VM's disk is just a file the hypervisor manages.

Virtual disks and datastores

A virtual disk is a file (or set of files) that a VM sees as a physical drive. Formats include VMDK (VMware), VHDX (Hyper-V) and qcow2 or raw (KVM/Proxmox). Because disks are files, they can be copied, moved and snapshotted easily.

A datastore is where the hypervisor keeps those files. Datastores can be built on several kinds of storage; see DAS, NAS and SAN for the underlying technology.

Shared storage, of any kind, is what allows high availability and live migration, because every host in the cluster can reach every VM's disks.

Thin and thick provisioning

With thin provisioning, a 200 GB virtual disk might use only 30 GB on the datastore until the guest writes more. This lets you overcommit storage, giving VMs more total capacity than physically exists. It is efficient but dangerous if many VMs grow at once: a full datastore can pause or crash every VM on it. Thick provisioning reserves all the space at creation, which is predictable but wastes capacity that is never used.

Tip: if you thin provision, set alerts at around 75 to 80 percent datastore usage, track the overcommit ratio and remember that snapshots and swap files also consume space. Arrays may thin provision too; thin on thin needs double monitoring.

Useful features and good practice

• Storage migration (such as Storage vMotion or Hyper-V storage migration) moves a running VM's disks to another datastore with no downtime, useful when retiring an array.
• Storage policies define requirements such as replication level or performance tier, and the platform places disks accordingly.
• Space reclamation (TRIM/UNMAP) returns freed blocks from the guest to the array so thin disks can shrink in real usage.
• Tiering and caching keep hot data on flash and cold data on cheaper disks.

When sizing, think about IOPS (input/output operations per second) and latency, not just capacity. A few busy databases can overwhelm a datastore with plenty of free space. Spread heavy workloads across datastores and monitor latency; sustained values above a few milliseconds on flash storage usually mean contention.

Finally, protect the data: storage features such as RAID and replication handle hardware failure, but you still need separate backups, ideally including an immutable or offline copy.