PRM Product Overview

7
You allocate:
• CPU resources in shares or processor sets
• Private memory and disk I/O bandwidth in shares
• Shared memory in megabytes
The following sections describe how these resources are managed by PRM.
Managing CPU allocation
PRM enables you to manage CPU resources at multiple granularities:
• Sub-core—FSS PRM groups are named for the PRM CPU scheduler FSS. An FSS PRM group can be
allocated resources of less than a single core or resources of multiple cores. The available CPU
resources are shared by all FSS PRM groups, with each group taking a portion of each core. (A
core is the actual data-processing engine within a processor. A single processor might have multiple
cores. Starting with HP-UX 11i v3, a core can support multiple execution threads. This feature is
known as Hyper-Threading. With Hyper-Threading disabled or unavailable, each core is seen as a
CPU. With Hyper-Threading enabled, each core can be seen as multiple, logical CPUs.)
• Whole-core—pSet PRM groups are based on processor sets (a collection of cores). The cores within
a pSet PRM group are dedicated to that PRM group. No processes outside the PRM group can
access those cores.
The PRM CPU scheduler takes precedence over the normal HP-UX CPU scheduler to ensure that FSS
PRM groups are allocated their specified share of CPU cycles. CPU time is divided among the PRM
groups according to the configured CPU shares. With shares, a group’s CPU allocation is determined
by dividing the group’s number of shares by the total number of shares assigned. For example, if a
group has five shares and 50 shares were assigned to all the groups in the policy, the group has
5/50 or 10% of the available CPU resources.
With a pSet PRM group, you assign the desired number of cores to the PRM group. These cores are
then available only to the applications and users assigned to the group, effectively creating a partition
on your system.
Within either type of PRM group, the standard HP-UX CPU scheduler is in effect. That is, all processes
within a particular group compete with each other for CPU time based on their standard scheduling
priorities.