Copyright by William Lloyd Bircher 2010 - The Laboratory for ...

3.5 Methodology for Power Modeling
With an understanding of system level events that are visible to the processor it is
possible to apply the iterative modeling process as depicted in Figure 3.5. This procedure
utilizes linear and polynomial regression techniques to build power models for individual
subsystems. The user identifies workloads which target a particular subsystem (cache,
system memory, disk) and performs regression modeling using performance events as
inputs. The model is then applied to a larger set of workloads to confirm accuracy and
the lack of outlier cases. Depending on the outcome, the process is repeated with
alternate performance events as inputs. Though an exhaustive search of performance
events can be performed, a rapid solution is found when events are selected with high
correlation to subsystem activity. Details of the modeling process in Figure 3.5 are listed
below.
1. Measure subsystem-level power using subset of workloads. Begin with simple,
easy-to-run workloads.
2. Confirm that Coefficient of Variation is greater than α for the chosen workload.
The simplest workloads often do not generate sufficient power variation for model
tuning. For example consider any of the cache-resident workloads in SPEC CPU 2000
which generate little or no activity in subsystems outside of the processor cores such as
memory. Tuning the model based on these low-variation workloads may cause the
process to include performance events that do not correlate well with power.
40