PRM Product Overview

8
Figure 2 illustrates a system using FSS PRM groups to ensure that applications receive a certain
amount of CPU resources.
Figure 2. PRM (FSS) CPU scheduler
By creating two FSS PRM groups based on the users and assigning 50% of the system CPU resources
to each, the shared CPU resource is partitioned according to the established policy.
Free CPU cycles are available for use by processes in any FSS PRM group. Additionally, if the CPU
resources are 100% busy but one or more FSS PRM groups are not using their CPU shares, the busy
FSS PRM groups get the idle groups’ CPU cycles in proportion to the busy groups’ shares. For
example, suppose there are three users, each with a dedicated PRM group, defined as follows:
• “CEO” has 30% of the CPU resources
• “Joe_Mktg” has 20% of the CPU resources
• “Lisa_Sales” has 50% of the CPU resources
If processes in Joe_Mktg are idle but processes in CEO and Lisa_Sales are using all of the available
CPU resources, then the CEO processes receive 37.5% of the CPU resources and processes in
Lisa_Sales receive 62.5% of the CPU resources. That is, the busy groups get the spare CPU cycles in
a 3:5 ratio (the ratio of their configured CPU allocations).
You can limit a group’s borrowing of unused resources with per-group capping, available starting
with HP-UX 11i v3. For more information, see the “Limiting maximum CPU consumption” section.
With pSet PRM groups, a group’s cores are available only for use by the processes in that group. The
following figure shows an HP-UX instance divided into three processor sets. Processes in PRM group A
have no access to the cores in pSet 1 or pSet 2. Using PRM private memory management, you can
also isolate the memory available to the PRM groups.