Australian cloud for business software — built and run in Sydney since 2002

NetLeverage
NetLeverage

Shared SAN, hyperconverged or local disks: choosing storage for a private cloud

Technical article
Hosting

About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.

The storage design underneath a private cloud decides three things a business notices: whether its servers restart automatically when a physical host fails, whether hosts can be maintained without downtime, and what it takes to add capacity later. A shared storage array and hyperconverged storage both let virtual machines restart elsewhere; local disks on each host do not, unless separate replication is added. Between the first two, the choice is mostly about how you want failures, maintenance and growth to work.

Why it matters

When a provider proposes a private cloud, storage is hard to change once virtual machines live on it. A business that chooses local disks to save money may discover that a host failure means restoring from backup rather than a short automatic restart. A business that chooses a large array may find that growth arrives in expensive steps. Asking the right questions before signing avoids both surprises.

How it works

Shared storage array (SAN or NAS)

All hosts connect to one central storage system over a dedicated network. Every host can see every virtual machine's disks. Broadcom's vSphere documentation describes a high availability cluster as hosts connected "to the shared storage where your virtual machine data resides". If a host fails, another host can start the affected virtual machines because their disks are still reachable.

The array has its own controllers, disks and firmware, and its own support contract. Adding capacity means adding shelves or disks to the array, within the limits of that model.

Hyperconverged storage

Each host contributes its own internal drives, and software combines them into one pool that every host can use. Broadcom states that when vSAN is activated it "aggregates the specified local storage disks available on the hosts into a single datastore shared by all hosts". Microsoft's equivalent, Storage Spaces Direct, lets you "combine internal storage drives on a cluster of physical servers (2 and up to 16) into a software-defined pool of storage", connected over Ethernet with "no special cable or storage fabric". Note the licensing condition on the same page: Storage Spaces Direct is "included in the Datacenter editions of Windows Server" 2022, 2019 and 2016, the editions that sentence names; the page also calls it a core technology of Azure Local. The page does not set out how Azure Local itself is licensed, so ask the provider which product, edition and licensing its quote assumes.

Data is protected by keeping copies or parity across hosts. Whether losing a host loses data depends on the storage policy, the number of hosts and how failure domains are set. On vSAN, Broadcom warns that "VMs can become inaccessible if storage policy is set to no data redundancy (FTT=0) and there is a single point of failure such as the host/disk/diskgroup", and that if the failed part cannot be brought back, "a restore of VMs from the backups will need to be performed". Ask what policy each virtual machine uses. That protection has to match the compute reserve. Broadcom says the number of failures tolerated in the vSAN rule set "must not be lower than the capacity that the vSphere HA admission control setting reserved". In plain terms: there is no point keeping spare processing power for two failed hosts if the storage can only survive one.

Local disks on each host

Each virtual machine lives on the disks of one host, which no other host can reach. This is simple and inexpensive, and performance is predictable. But if that host fails, its virtual machines stay down until the host is repaired or the machines are restored elsewhere from backup, unless a separate replication product keeps copies on another host. The vSphere pages cited above describe restart only for storage every host can reach.

What to choose and why

The table is our editorial summary, not taken from a source; individual products differ.

Question Shared array Hyperconverged Local disks
Automatic restart after a host failure Yes, if the cluster is configured for it Yes, if the cluster is configured for it No, without separate replication
Single component that can stop everything The array: redundant controllers and power remove some single points of failure, but the array as a whole (its firmware, its enclosure, its site) remains one shared failure domain Spread across hosts, if the storage policy tolerates host failures Each host affects only its own machines
Adding storage Add to the array Add drives or hosts Add drives to a host
Adding compute without storage Add hosts Depends on the design; adding a host may add storage too Add hosts
Separate support contract for storage Possibly, for the array Part of the host platform Part of each host
  • Choose a shared array when you already need large storage that grows faster than processing, or when you have workloads such as large file archives that do not fit neatly into hosts.
  • Choose hyperconverged storage when compute and storage grow roughly together and you want one platform, one support path and growth by adding hosts.
  • Choose local disks only for workloads that can be down for the time it takes to restore them, or that protect themselves by running copies on several hosts.

A hypothetical example

A professional services firm plans a three-host private cloud for about 30 virtual machines, mostly application and database servers, with steady growth. Hyperconverged storage would let it grow by adding a fourth host when both processing and storage approach their limits. A 20-terabyte document archive that grows faster than anything else might sit better on separate file storage than force early host purchases. The figures are illustrative.

Questions to ask the provider

  1. If one host fails, which virtual machines restart automatically and how long do they take?
  2. If the storage controller or one storage node fails, what keeps running?
  3. How many host failures can the storage survive, and does the compute reserve match?
  4. What is the next capacity step, and what does it cost?
  5. Can a host be taken offline for maintenance without stopping virtual machines?

Risks and limits

  • Redundant on paper, not in practice. An array with a single controller, or a hyperconverged cluster set to tolerate fewer failures than the reserve assumes, can still stop everything. Component redundancy inside one array does not protect against losing the whole array or the site; that needs a second copy elsewhere.
  • Rebuild periods. In general terms (our summary, not a quoted source), after a drive or host failure, storage systems copy data to restore protection; performance and protection are reduced until that finishes.
  • Growth coupling. As a general design point (our summary), hyperconverged designs tie storage growth to hosts; arrays tie it to the array's own limits.
  • High availability is not backup. None of these designs protects against deletion, corruption or ransomware; backups are a separate decision.
  • Product specifics change. Check current documentation for the exact product and version proposed.

Sources

Opened 15 November 2026.

Related articles: How to build and validate a VMware private-cloud cluster, How to design and validate vSAN storage policies and maintenance headroom and vSphere HA did not restart some VMs go deeper on the VMware side.

Related service: Cloud hosting.

Learning notes

  • Storage design decides whether virtual machines can restart on another host.
  • Shared arrays centralise disks; hyperconverged storage pools each host's internal drives.
  • Storage failure tolerance must match the processing reserve, or the reserve is wasted.
  • Local disks are simple but leave a failed host's machines down until repaired or restored.
  • High availability does not replace backups.