The Oracle Partitioning Policy is an Oracle document that states when a customer may license fewer Oracle Processor licenses than the total physical cores of a server. It classifies virtualization and partitioning technologies as soft partitioning, hard partitioning or an Oracle Trusted Partition, and sets the conditions under which each may be used “to license a sub-capacity of total physical cores as an exception from the contractual Oracle Processor definition”.[1] The current edition gives guidelines on policies in effect as of 14 February 2022. It also covers Capacity on Demand servers and Oracle programs running in containers and Kubernetes clusters.
The document describes itself as educational. It “may not be incorporated into any contract and does not constitute a contract or a commitment to any specific terms”, and Oracle may modify its definitions and conditions from time to time.[1] The contractual starting point is the Processor definition in the License Definitions and Rules: “all processors where the Oracle Programs are installed and/or running”.[2] The partitioning policy states the exceptions Oracle recognises to that definition.
Partition types
The policy defines partitioning as separating the CPUs on a server into sections that each act as a separate system. It distinguishes three categories.[1]
| Type | How the policy describes it | Effect on Processor licensing | Catalog |
|---|---|---|---|
| Soft partitioning | The operating system uses resource managers to limit the CPUs where a program runs; capacity can be changed easily. Examples: Solaris 9 Resource Containers, AIX Workload Manager, HP Process Resource Manager, Affinity Management, Oracle VM and VMware | Not permitted to determine or limit the licenses required for a server or cluster, unless the policy states otherwise | Hard partitioning / soft partitioning |
| Hard partitioning | Physically segments a server into self-contained systems with their own CPUs, operating system, boot area, memory, I/O and network resources | Only the listed technologies, each capped at a maximum number of cores or processors, may limit licenses | Approved hard partitions must be capped |
| Oracle Trusted Partitions | Oracle VM Server or Oracle Linux KVM on approved Oracle Engineered Systems, monitored by Oracle Enterprise Manager | Sub-capacity licensing at 2 vCPUs per physical core, with a high-water mark | Oracle Trusted Partitions |
The approved hard partitioning technologies are:[1]
- Physical Domains (PDomains, Dynamic Domains or Dynamic System Domains).
- Solaris Zones and Solaris Containers (capped only).
- IBM LPAR, including DLPAR with AIX 5.2.
- IBM Micro-Partitions (capped only).
- vPar (capped only) and nPar.
- Integrity Virtual Machine (capped only).
- Secure Resource Partitions (capped only).
- Fujitsu PPAR.
The policy states that “no other technology or configuration” qualifies.
Metrics
The policy affects licenses counted on the Processor metric. Under that metric the licensed cores are multiplied by the core factor from the Oracle Processor Core Factor Table, with fractions rounded up (Processor Core Factor).[2] The Core Factor Table lists, for example, 1.0 for IBM POWER8, POWER9 and POWER10 and for “All Other Multicore chips”, and 0.5 for the AMD EPYC and Intel Xeon families it names.[3] The policy also matters for Named User Plus licensing. Named User Plus minimums such as 25 per Processor for Database Enterprise Edition (Named User Plus minimum) are calculated from the processor count, so a partition that is not recognised raises the minimum as well.[2]
Counting / floors
Soft partitioning and clusters
Where only soft partitioning is in place, the policy gives no basis for counting fewer than the physical processors of the server, or of the cluster of servers, on which the program is installed or running.[1] The policy names VMware and Oracle VM among its soft partitioning examples. It does not describe how VMware clusters, vCenter instances or storage should be scoped. Statements about those scopes do not come from this document.
Hard partitioning conditions
Every approved hard partitioning technology must have “a capped or a maximum number of cores/processors for the given partition”. The policy adds specific exclusions (IBM Live Partition Mobility and TurboCore are not sub-capacity):[1]
- IBM processors in TurboCore mode cannot be used to reduce licenses; all cores must be licensed.
- IBM PowerVM Live Partition Mobility is not approved hard partitioning. All cores on both the source and the destination servers must be licensed.
Oracle’s own hypervisors qualify only under stated configurations. Oracle Linux KVM, Oracle VM Server for x86 and Oracle VM Server for SPARC count as hard partitioning only if specific cores are allocated as set out in separate Oracle configuration documents. Oracle Solaris Zones qualify only as described in their own hard partitioning document (Oracle Linux KVM and Oracle VM as hard partitioning).[1]
Oracle Trusted Partitions
On approved Oracle Engineered Systems (Exalogic Elastic Cloud, Exalytics In-Memory Machine, Private Cloud Appliance and Exadata Database Machine, each with a minimum software version), Oracle permits Oracle VM Server or Oracle Linux KVM to limit Processor licenses. The virtual machines must be monitored by Oracle Enterprise Manager, except on Private Cloud Appliance 3.01 and later. The counting rules are (Trusted Partitions: 2 vCPUs per core, increments of 2 cores):[1]
- two virtual CPUs count as one physical core;
- licenses are procured in increments of two physical cores;
- the customer licenses the highest number of vCPUs running at any time (the high-water mark), but never more than the total physical cores in the machine;
- hyper-threading must be enabled;
- the minimum is two physical cores, except on Exadata, which has its own minimum in the Exadata licensing guide.
As an illustration, Trusted Partitions on a Private Cloud Appliance whose largest concurrent footprint for a program is 12 vCPUs count as 6 physical cores. A footprint of 13 vCPUs counts as 8 physical cores, because licenses are bought in increments of two cores. The policy expresses the result in physical cores and does not itself restate the core-factor step of the Processor definition.[1][2]
Capacity on Demand
Oracle allows customers to license only the cores that are activated when a server is shipped (“Capacity on Demand” or “Pay as You Grow”). Oracle offers no special terms for “Pay Per Use” or “Pay Per Forecast” models, where the number of CPUs can be scaled down or varied (Capacity on Demand: license activated cores).[1]
Failover and backup testing
Two contractual rights in the License Definitions and Rules sit alongside the partitioning policy. Under the failover right, a licensed program may run on one unlicensed spare node in a clustered environment that shares a logical disk array in one data center. This is allowed for up to ten separate 24-hour periods per calendar year, and maintenance downtime counts towards the limit (Failover: ten 24-hour periods per year on an unlicensed spare). Under the testing right, an Oracle Database license may be run on an unlicensed computer to test physical copies of backups up to four times a year, for no more than two days each time. Remote mirroring and other methods that copy or synchronise the binaries are not covered (Testing physical copies of backups on an unlicensed computer).[2]
Containers and Kubernetes
Once a container image that contains Oracle programs has been pulled to a host or a Kubernetes node, the host or node must be licensed for the processors on it. For a physical host that is all of its processors. For a virtual machine the number of processors follows the partitioning rules above (Containers and Kubernetes: license every host where the image is pulled).[1] This section does not apply to Solaris Containers, Solaris Zones or Windows Containers. For limiting the nodes of a Kubernetes cluster, the policy refers to a separate Oracle document on running and licensing programs in containers and Kubernetes. The Technology Global Price List applies the same approach to Verrazzano Enterprise Container Platform: all processors in the Kubernetes nodes where Verrazzano images are pulled must be counted, and virtual nodes follow the Partitioning Policy.[5]
Cloud / BYOL
Public cloud counting is not part of the partitioning policy. Oracle covers Amazon EC2 and RDS, Microsoft Azure and Google Cloud Platform in the separate Authorized Cloud Environments policy, whose edition as of 4 September 2026 counts two vCPUs as one Processor license where multi-threading is enabled. The Core Factor Table does not apply in those environments (ACE vCPU to Processor).[4] The same policy caps Standard Edition 2 at eight vCPUs per instance and Standard Edition at 16 vCPUs (ACE policy 2026: Standard Edition caps and Oracle Linux counting). Oracle Cloud Infrastructure database services use their own BYOL conversion ratios. See Oracle Cloud Infrastructure BYOL and Universal Credits.
Programs
The partitioning policy names no commercial program. Processor counts for unlimited license agreements are fixed at certification under the terms of each agreement. The Authorized Cloud Environments policy excludes licenses deployed in those clouds from ULA certification (ULA licenses in ACE).[4] See Oracle ULA and PULA.
Out of scope
- The Oracle Partitioning database option, a separately priced Enterprise Edition option for table and index partitioning, which is covered in Oracle Database options and management packs.
- Hypervisors not named in the policy. The document neither approves nor specifically discusses them, beyond the general soft partitioning rule.
- Oracle VM VirtualBox, which is licensed on its own Socket and Named Workstation User metrics.
- Oracle Linux and Oracle VM support subscriptions, which use separate metric definitions.