HP-UX Workload Manager User's Guide

Configuring WLM
Defining the PRM components (optional)
Chapter 5182
larger than the requests, the requests are used first. Thus, A, B, and C all
receive a CPU allocation of 20. That leaves 40% of the CPU resources
unallocated. With distribute_excess not set, all 40% goes to OTHERS.
Table 5-3 shows the same example, but with distribute_excess set.
(This discussion assumes extended_shares is not set.) Again, A, B, and
C all receive a CPU allocation of 20, leaving 40% of the CPU unallocated.
This time though, distribute_excess is set. Consequently, the
remaining 40% (less the 1% default to OTHERS) goes to A, B, and C—based
on their weights. Dividing the weight of A (2), by the sum of the weights
(4), A gets 50% of the total CPU resources, while B and C get 25%. WLM
removes 1% from B to give to OTHERS so that it maintains the minimum
CPU percentage required for all groups.
Table 5-2 weight example with distribute_excess = 0 (off)
Group Weight
Requested
allocation
Final
allocation
A 220 20
B 120 20
C 120 20
OTHERS 1 (default) 1 (default) 40
Table 5-3 weight example with distribute_excess = 1 (on)
Group Weight
Requested
allocation
Final
allocation
A 22050
B 12024
C 12025
OTHERS 1 (default) 1 (default) 1