A Resource Utilization Score for Software Energy Consumption

Abstract

Abstract-Software as the true consumer of power and its potential contribution to reach sustainability goals is increasingly being acknowledged. Studies so far have presented successful results and methods to address the energy consumption of the software, indicating that different stakeholders striving for green software have different information needs with respect to their goals. However, currently there is no uniform manner to communicate measurements to the different stakeholders such that key findings are clearly identifiable and easy to understand, which is likely to hamper green software practices. In this paper we propose a metric that expresses a score for the resource utilization, such as power consumption, of a software product. The metric is designed to be a single score and is flexible to encompass those aspects that a stakeholder considers relevant in the context of software energy consumption. The metric was applied on two applications and allowed for objective comparison of application configurations and versions. Also the behavior of these applications across different hardware configurations could be analyzed. In addition to the metric we investigate means to visualize measurements which enhances communication and helped with highlighting the key findings.

The hunt for the guzzler: Architecture-based energy profiling using stubs

“…In turn, the metrics, of which a solid list is available in literature [28], determine the measurement approach that a stakeholder should apply. Ideally, since different hardware could show different energy consuming behavior [21], different configurations are used to validate the energy profile. However, as this increases the costs, the minimal requirement is to have one environment that is representative for the production environment, where representative means using hardware and a software stack that closely resemble the target environment for the software.…”

“…Different implementations potentially impact the energy consumed by the software, thus negating the possibility to generalize findings across products. In addition, different hardware configurations can significantly impact the effect of a specific implementation [21]. Hence, each product would still require its own analysis when energy concerns are addressed at code level.…”

“…To address qualitative concerns during software design requires a more predictive approach (e.g., [19]) and validated design decisions (e.g., [20]). However, as the actual energy consumption differs per configuration of both hardware and software [21], the end result of sustainable design efforts still requires verification. With the StEP Method we focus on analyzing a single version of a software product during development, to pinpoint the energy consuming elements within a single release.…”

“…The performance of the test server was measured using Windows Performance monitor, a standard tool that is part of the Windows operating system. Following [21], the relevant performance metrics were determined with the product owner:…”

“…While the generalization of the study could be improved by diversifying the hardware involved in the experiment, obliging stakeholders to perform measurements on multiple hardware environments would significantly increase the required investments for applying the StEP Method. And even if such an effort is made, the generalizability of the results across hardware platforms is expected to be limited [21]. Hence, to obtain the best energy consumption measurements for CITEX, we chose to perform the experiment in an environment that closely resembles the actual production environment.…”