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

NetLeverage
NetLeverage

Adding CPU, memory or disk to a hosted server: which changes need a restart

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.

On a VMware-hosted server, a disk can be made bigger while the server runs only when specific conditions are met: a supported disk controller and vSphere version, no snapshots, and a disk that is not shared between machines. Some guest operating systems also need a reboot to see the new size, and one ESXi 8.0.2 defect means the provider must check its patch level before moving a disk it has grown live. Volumes holding SQL Server databases need a short planned window regardless. Adding processors or memory while the server runs is possible only if live-add was switched on for that virtual machine beforehand, the platform licence includes it and the operating system supports it. Otherwise it needs a shutdown. Switching on memory live-add also has a performance cost on larger servers, so it is a choice to make deliberately, not a default to tick.

Why it matters

The difference decides whether an upgrade happens at 2 pm on a Tuesday or in a booked window at night. If you assume every change is live and it is not, users are interrupted. If you assume every change needs a restart, you may wait days for something that could have been done in minutes. And if live-add is switched on for every server "just in case", some larger servers may run slower all the time.

How it works

Disk space: two steps, with conditions

  1. The provider enlarges the virtual disk on the platform. Broadcom's knowledge article on growing virtual disks sets the conditions: "Do not alter the base disk in any way without committing all snapshots first or they will be invalidated", and "Some virtual machine guest operating systems may require a reboot to detect changes to the size of the virtual disks". Growing the disk while the server runs (Broadcom calls it hot extend) has its own conditions. Broadcom's article on disks that cannot be enlarged says "PVSCSI controller allows hot extend disk from vSphere 7.0. NVMe controller allows hot extend disk from vSphere 8.0.1." and "All snapshots must have been removed." A disk shared between machines cannot be grown live: "VMware vSphere does not support Hot Extend (online expansion) for virtual disks with the Multi-writer flag enabled." On ESXi 8.0.2, Broadcom warns that if a disk is grown live and the machine is then moved to other storage, "data written to the growing area by the guest may not get copied"; it states the issue "has been fixed in 8.0.2 P03". Ask the provider which controller your disks use, whether its hosts are patched and how it handles a live grow.
  2. Someone extends the volume inside Windows so the extra space is usable. Microsoft explains that you add space by "extending the volume into available unallocated (empty) space on a disk drive".

The disk layout details (what has to sit next to the volume, and why shrinking is a different matter) are in our companion article on growing a hosted disk.

Memory: live only if prepared

Broadcom's knowledge article on memory hot add lists the conditions. The host needs the right licence feature: "Verify if you have Hot Pluggable Virtual Hardware". The virtual machine needs the option enabled. And "it is equally important that the guest operating system supports the hot add feature". The same article names two more blockers: "If the setting memoryReservationLockedToMax is set to true, hot-add will fail", and "If the configured Guest OS in VM settings does not match the actual installed OS, the hot-add option may be greyed out or fail".

There is a catch. Large virtual machines benefit from vNUMA, which tells the operating system how memory is laid out across the physical processors so it can keep work close to its memory. Broadcom states that in vSphere 8 "Memory Hot Add still completely deactivates vNUMA", which "can impact the performance of large virtual machines (typically those with 8 or more vCPUs)".

Processors: improved in vSphere 8

Processor live-add has the same preparation requirement. Broadcom's rightsizing guidance for ESXi 8.0 says the newer virtual hardware "preserves vNUMA when CPU Hot Add is enabled", and notes "This is a change from earlier vSphere releases, where activating CPU Hot Add disabled vNUMA for the VM." It sets two conditions for that: "The VM must be running virtual hardware version 20 or later. Cores per Socket must be set to "Assigned at power on"". So where both are met, enabling processor live-add has less downside than it once did. On older virtual hardware, the old trade-off may still apply. Our published article One vNUMA node, yet CPU Hot Add blocks VM reconfiguration covers a case where it still gets in the way.

Without preparation

If live-add was not enabled for a resource, changing that resource means shutting down the virtual machine, changing it, and starting it again. Our advice is to treat switching live-add on as a change for a planned restart, decided when the server is built; confirm the exact steps with your provider.

What to choose and why

Change Live? Conditions
Enlarge a disk and extend the volume Only if the conditions are met Supported controller (PVSCSI from vSphere 7.0, NVMe from 8.0.1); no snapshots; not a multi-writer disk; ESXi 8.0.2 patched to P03 before a later storage move; some guests need a reboot to see the size; free space directly after the volume inside Windows; SQL Server volumes need a backup and service stop
Add memory Only if prepared Licence feature, option enabled, guest support, memory reservation not locked to maximum, guest OS setting matches; disables vNUMA
Add processors Only if prepared Licence feature, option enabled, guest support; vNUMA kept on hardware version 20 or later with Cores per Socket set to "Assigned at power on"
Remove processors or memory Plan for a restart Ask the provider; do not assume live removal

The table is our editorial summary of the Broadcom and Microsoft sources above for VMware vSphere 8; it is not a guarantee for any particular plan.

A sensible approach, as our editorial advice:

  • Any server where you want memory live-add: ask the provider to assess what losing vNUMA would mean for that server, and to measure it where it matters. Broadcom's "typically 8 or more vCPUs" is a guide to where the effect usually shows, not a safe limit; the real impact depends on the memory size, the physical host's layout, the guest operating system and the workload.
  • Large servers and databases: unless that assessment says otherwise, leave memory live-add off, size memory with headroom, and book a window for growth.
  • All servers: build disks with free space directly after the data volume, so extending stays simple.

A hypothetical example

An accounting firm's file server runs out of space at month-end. The server has no snapshots, its disk sits on a PVSCSI controller and the provider's hosts are patched, so the provider enlarges the disk during the day and the firm's IT contact extends the volume in Disk Management without a restart. The same week, the practice-management database, a 12-processor server, needs more memory. Live-add was left off on purpose, so the change is booked for 7 pm and takes a single restart. The example is illustrative.

Risks and limits

  • Not all plans include live-add. It depends on the provider's licensing and policy; ask.
  • The operating system must cooperate. Check the guest operating system's support, and expect the vendor of your application to have opinions too.
  • Licensing inside the server may change. Adding processors can affect licences counted per core; check before you add.
  • Snapshots, shared disks and host patch level can block or complicate live disk growth. The provider checks these before it grows a disk live.
  • Database volumes are different. For a volume holding SQL Server database files, Microsoft advises that "you should back up all system and user databases and stop SQL Server services first", so treat that growth as a planned window, not a live change.
  • Other hypervisors differ. This article covers VMware vSphere 8; Hyper-V and others have their own rules.

Sources

Opened 15 November 2026; Broadcom articles reopened 16 November 2026.

Related articles: One vNUMA node, yet CPU Hot Add blocks VM reconfiguration; Windows Disk Full, VMware Space Free: Map the Layers.

Related service: Cloud hosting.

Learning notes

  • A disk can be enlarged while the server runs only with a supported controller, no snapshots and no multi-writer flag; some guests still need a reboot to see the new size.
  • Memory and processor live-add need the licence feature, the option enabled beforehand and guest support.
  • In vSphere 8, memory live-add disables vNUMA; Broadcom says the effect is typically on machines with 8 or more vCPUs, but ask for an assessment rather than relying on the count. Processor live-add on newer virtual hardware keeps vNUMA.
  • Decide on live-add at build time or the next planned restart.
  • Database volumes need a backup and a service stop before they are grown.