Server virtualisation is mature, well understood and almost universal. It saves hardware spend, floor space, power and time, which is why practically every large organisation runs it somewhere.
It also changes what you owe on software, and that part is routinely overlooked. Unless you understand the implications and can see your licence position as the environment moves, you can end up paying more in additional licences — and in settlements, if a vendor finds the shortfall first — than you ever saved by virtualising.
Why the rules change
Most publishers licence physical and virtual environments differently. The specifics vary enormously, but one theme is consistent across all of them: small changes to the virtual environment can have a large effect on what you are required to hold.
The two usual triggers are:
- Hardware. Additional hosts or additional CPUs in a cluster.
- Mobility. Server mobility within a farm — DRS, vMotion or the equivalent — which can make every host in the cluster licensable rather than just the one the workload happens to be running on.
Converting a physical machine to a virtual one is not a like-for-like move in licensing terms. It changes the basis on which the product is counted, and the only place to find out how is your own agreement.
Maintenance matters too. Several publishers make specific rights conditional on active maintenance or Software Assurance — Microsoft’s licence mobility rights for server applications within a farm are the example most organisations meet first. Whether that applies to you depends on the product and the agreement you signed, so it is worth confirming rather than assuming.
The questions to ask before you virtualise
Can you track the movement of virtual servers across the farm?
In an audit you may be asked to demonstrate that a workload has not moved somewhere that would have made additional hosts licensable. Assertion will not do it; you need the record.
Oracle in particular documents what it recognises as hard and soft partitioning, and the distinction decides whether you can legitimately limit a workload to a subset of a cluster or whether the whole cluster counts. Understanding which side of that line your configuration falls on is not optional if you run Oracle on VMware.
What ships today. This is the question the original version of this article left with the reader, and it no longer needs to be. CerteroX ITAM and CerteroX SAM connect directly to the virtualisation layer — VMware, Microsoft Hyper-V, Citrix XenServer, IBM HMC, Oracle VM, Red Hat oVirt and Nutanix — so host, cluster and guest relationships are inventoried rather than described. On top of that:
- The Oracle engine models processor types and core factors, options and packs with evidence and override, licence pools with hosting rights and geographic rules, cover-down logic for Enterprise Edition and uncapped quantity for unlimited agreements. That is the level of detail an Oracle LMS engagement actually turns on.
- IBM sub-capacity is handled with PVU and Virtual Processor Core metrics, an ILMT connector with compliance gap analysis, and enforcement of the 30-minute inventory cycle that sub-capacity licensing genuinely requires. Miss that cycle and you are licensed at full capacity whatever the hypervisor says.
- Microsoft server licensing covers device and user CALs, named user and external connectors, and SQL Server and Windows Server core and processor licensing with cluster and virtualisation awareness.
- Access Control rules cover RDS, Citrix and VDI streamed-application licensing, and non-persistent VDI is inventoried as a first-class case rather than producing a new machine record every morning.
Do you consider licensing when you design or upgrade a farm?
Cluster design is usually settled on capacity, resilience and cost of hardware. The licensing consequence arrives months later as a surprise.
Knowing the effect on the specific products you intend to run turns that into an input rather than an outcome. Sometimes it argues for a smaller, dedicated cluster. Sometimes it argues for pinning a workload. Occasionally it argues for not virtualising a particular product at all. All three are cheaper decided at design time.
Do you consider licensing when you deploy a new operating system environment into the farm?
Spinning up another OSE is a two-minute operation and can be a five-figure one. The obvious example is Windows Server: do you have the CALs, and are you licensing by core across every physical core in the host, with the minimums the product terms impose?
Windows Server has been licensed per physical core since Windows Server 2016, not per processor. If your mental model of this predates that change — and many do, because the change is old enough to have been absorbed without ever being re-examined — it is worth re-checking your cluster maths from scratch.
Is the farm a dedicated test or development environment?
Development and test workloads often carry different rights, but the conditions attached to them are strict and vary by publisher — who may use them, on what, and whether any production data may touch them. Treating a cluster as “just dev” without checking those conditions is one of the more common findings in an audit.
CerteroX SAM handles this with the Exclude From Licensing workflow for MSDN, development, training and second-use devices, so those machines are inventoried and visible but do not distort the position — and the exclusion is recorded, with an audit trail, rather than applied silently.
SQL Server: which model?
SQL Server can be licensed several ways in a virtual environment — per virtual core on the guests, or per physical core on the host with virtualisation rights, with the answer depending on density, edition and whether you hold Software Assurance. The right choice is genuinely organisation-specific, and the difference between the best and worst option on a busy cluster is large.
You cannot make that comparison without knowing your actual density: how many guests, how many virtual cores, on how many physical cores, running which editions. That is an inventory question before it is a licensing question.
The underlying point
Virtualisation did not make licensing harder because publishers made the rules more complicated. It made licensing harder because it decoupled the software from the hardware, and almost every licence metric ever written counts hardware.
That means your licence position is now a function of a configuration that changes continuously, often automatically, and frequently without anyone thinking about licensing at the moment it changes. A position calculated once a year is a position that was correct once a year.
The answer is to compute it continuously against live inventory from the hypervisor, rather than reconstructing it from memory when a letter arrives.
To see a virtual cluster modelled against the publisher rules that actually apply to it, book a demo.