On the energy consumption and performance of systems software

Abstract

Models of energy consumption and performance are necessary to understand and identify system behavior, prior to designing advanced controls that can balance out performance and energy use. This paper considers the energy consumption and performance of servers running a relatively simple file-compression workload. We found that standard techniques for system identification do not produce acceptable models of energy consumption and performance, due to the intricate interplay between the discrete nature of software and the continuous nature of energy and performance. This motivated us to perform a detailed empirical study of the energy consumption and performance of this system with varying compression algorithms and compression levels, file types, persistent storage media, CPU DVFS levels, and disk I/O schedulers. Our results identify and illustrate factors that complicate the system's energy consumption and performance, including nonlinearity, instability, and multi-dimensionality. Our results provide a basis for future work on modeling energy consumption and performance to support principled design of controllable energy-aware systems.

Model discovery for energy-aware computing systems: An experimental evaluation

“…As we have shown [12], even simple systems that de/compress files exhibit complex performance and energy profiles. Control theory holds great promise for the automated and optimal control of such energy-aware systems.…”

“…Based on the extensive data set we collected in [12], we plan to evaluate additional MIMO models, especially those with more than two inputs/outputs, and those that explore other dimensions (e.g., different compression algorithms and compression levels). We also plan to design and implement actual SISO/MIMO controllers and evaluate their ability to trade off power consumption and performance.…”

“…We recently concluded an extensive year-long study [12] of the energy consumption and performance of a file-compression server, a representative workload involving both substantial CPU usage and disk I/O. We analyzed the effects of several input parameters, including compression algorithm, compression level, file type, persistent storage media, CPU Dynamic Voltage and Frequency Scaling (DVFS) level, and disk I/O scheduler-all under the Linux operating system.…”

“…In this paper, we present a model-discovery methodology that mitigates the complexities we identified in [12] to achieve accurate plant models of energy-aware systems. Key to our approach is to use only numeric model parameters as inputs and outputs, and treat non-numeric model parameters as part of the workload.…”

Software energy consumption prediction using software code metrics

“…Regarding optimization, a primary goal of energy consumption measurements is to optimize the system for reduced energy consumption. Li et al [108] applied the idea of load balancing to server-room heating and cooling and Fei et al [55] used context-aware source code transformations and achieved power consumption reductions of up to 23% for their software. Selby [167] investigated methods of power reduction by code transformations and compiler optimizations.…”