Full-System Power Analysis and Modeling

The increasing costs of power delivery and cooling, as well as the trend toward higher-density computer systems, have created a growing demand for better power management in server environments. Despite the increasing interest in this issue, little work has been done in quantitatively understanding power consumption trends and developing models to predict full-system power and component power breakdown that are not hardware specific.
To quantitatively understand power consumption trends in server systems, we study the component-level power breakdown and variation, as well as temporal workload-specific power consumption of two instrumented systems: a power-optimized blade system and a compute-optimized Itanium-2 system. Using this analysis, we examine the validity of prior ad-hoc approaches to understanding power breakdown and quantify several interesting trends important for power modeling and management.
We also introduce Mantis, a non-intrusive method for modeling full-system power consumption and providing real-time power prediction. Mantis is not hardware-specific and uses a one-time calibration phase to generate a model correlating AC power measurements with standard user-level system utilization metrics.
We experimentally validate Mantis on the two instrumented systems using a variety of workloads (CPU, memory, or I/O intensive) and hardware configurations (number of processors, power supply and frequency settings). Mantis provides power estimates with high accuracy for both overall and temporal power consumption, making it a valuable non-hardware-specific tool for power-aware scheduling and analysis.