HP-UX Workload Manager User's Guide

Example configuration files
distribute_excess.wlm
Chapter 9288
# tune structure
#
# We must define a data collector for every metric used in
# SLO definitions. In this example, we add a global tune structure
# where we enable distribution of excess shares to groups defined in
# the prm structure above. This is accomplished by simply setting
# the keyword distribute_excess to 1 (TRUE). Without this setting,
# HP-UX WLM would allocate all excess shares to the OTHERS group.
#
# HP-UX WLM will, by default, provide an allocation sufficient to meet
# the specified goals. Let us assume that both goals are being met and
# that the Orders group is consuming 20% of the CPU resources and
# the Sales group is also consuming 20%. Then 60% of the machine is
# available for distribution to the two groups. If distribute_excess
# is enabled, WLM will ultimately allocate about 25% of the CPU
# resources to the Orders group and about 75% of the CPU resources to
# the Sales group, reflecting the 3:1 weight ratio specified for the
# groups in the prm structure. (The OTHERS group must get a minimum
# of 1% of the CPU. This 1% will be taken from either the Sales group
# or Orders group. As a result, WLM will not be able to grant CPU
# exactly in the 3:1 weight ratio.)
#
# Enabling distribute_excess is likely to reduce the response
# times for both groups well below the goals specified (for example,
# the sales_request_time may drop down to 3 seconds). Without the
# distribute_excess keyword set, HP-UX WLM would keep the response
# times close to the goals (so, sales_request_time would be kept
# around 4.5 seconds).
#
# With distribute_excess enabled, HP-UX WLM allocates excess shares so
# that the resulting allocations are as close as possible to the
# weight distribution without decreasing allocations resulting from
# SLOs. Thus, HP-UX WLM will distribute CPU resources sufficient to
# meet the SLOs and only after that action will HP-UX WLM begin
# distributing excess shares to achieve the weight distribution.
#
# Consider yet another scenario where SLO goals are being met.
# However, the Orders group is using 50% of the resources to meet its
# response time goal, and the Sales group is using only 20%. HP-UX WLM
# tries to match the allocations according to the weight
# specifications, but it will not reduce an allocation if doing so
# will cause the SLOs to fail (the goals to not be achieved).
# HP-UX WLM will give the Sales group the remaining 30% of the CPU
# resources so that the resulting allocation is 50% for each group.
# This does not meet the 3:1 ratio (that is, 75%/25%), but doing so
# would cause the order_processing SLO to fail. (As OTHERS must get