Micro-Partitioning
From Wikipedia, the free encyclopedia
Micro-Partitioning, also known as shared processor partitioning,[1] is a form of logical partitioning which was introduced by IBM on systems using the POWER5 processor. It only differs from a dedicated processor partition in the way CPU utilization is configured and managed by the POWER Hypervisor (PHYP) firmware.[2] All IBM POWER5 and POWER6 systems are partitioned and will run "on top" of the PHYP.[3]
The POWER Hypervisor controls time slicing, management of all hardware interrupts, dynamic movement of resources across multiple operating systems, and dispatching of logical partition workloads.
When a shared processor partition is activated by the PHYP, the logical partition (LPAR) is guaranteed a certain processing capacity, if needed, and a number of virtual processors, based on configuration and current availability. The processing capacity is drawn from a pool of shared processor resources.[4]: 6
The minimum processing capacity per processor is 1/10 of a physical processor core, with a further granularity of 1/100.[5][6][7] The PHYP uses a 10 ms time slicing dispatch window for scheduling all shared processor partitions' virtual processor queues to the PHYP physical processor core queues. A shared processor partition can be either capped or uncapped. A capped partition can never exceed the currently configured processing capacity, whereas an uncapped partition can exceed the currently configured processing capacity up to 100% of the number of the currently configured virtual processors.[4]: 10
If the shared processor partition is DLPAR capable, the number of virtual processors and processing capacity can be altered dynamically for the partition.[4]: 9