Resizing partitions automatically with HP-UX Workload Manager

# SLO (Service-Level Objective) structure.

slo slo_Orders {

# Priority of this SLO.

pri = 1;

# The workload group this SLO applies to.

entity = PRM group Orders;

# Request no less than mincpu of available shares.

mincpu = 1;

# Grant no more than maxcpu shares for this SLO.

maxcpu = 12800;

# Allocate available CPU shares to achieve goal.

goal = usage _CPU;

}

# Global tune structure is used to set values to global tune variables.

tune {

# Set interpretation of share:

# 1 => share is 1/100 of one CPU

# 0 => share is 1/100 of all CPUs

# Do not set to 0 if using PSET groups.

absolute_cpu_units = 1;

# Set rate (seconds) at which WLM changes resource allocations:

wlm_interval = 5;

}

You might have noticed the line with absolute_cpu_units = 1. When using partitions, WLM

requires SLOs to use absolute CPU shares, which is what happens when absolute_cpu_units

= 1. (When not using absolute CPU shares, a share is 1% of the total CPU resources on a system.

In a partition, the value represented by 1% of the CPU resources fluctuates as the number of cores

assigned to the partition changes. Thus, when managing partitions, a different share definition is

used to maintain a constant share value: one share is 1/100 of a single core.)

d. Create a WLM configuration (using the wizard) on each partition that will be managed by the

same global arbiter instance.

e. Start the WLM daemon (wlmd) on each partition: # wlmd –a config_file.

4. Set up the WLM global arbiter in the “primary host” partition as described in the following section,

“Configuring automatic partition resizing.”

Configuring automatic partition resizing

After creating and activating WLM configuration files in each partition on the system, as explained in

“Configuring HP-UX Workload Manager on a partition without isolating the partition’s workloads” (on

page 11) and “Configuring HP-UX Workload Manager to isolate the workloads within the partition”

(on page 12), you must configure the WLM global arbiter to move cores among the partitions. To set

up the WLM global arbiter:

1. Log in to the partition where wlmpard will run (the partition identified by primary_host in all the

WLM configuration files). This partition will have both a WLM configuration file and a WLM

global arbiter configuration file. (Technically, the global arbiter configuration can be run on any

system with WLM installed and with network connectivity to the managed partitions. However,

using one of the managed partitions as the primary host simplifies this example.)

2. Create the wlmpard configuration file. It has a single structure, the par structure, which indicates,

for example, the interval (in seconds) for how frequently to move cores among partitions, the