On reducing the energy consumption of software product lines

Abstract

Along the last decade, several studies considered green software design as a key development concern to improve the energy efficiency of software. Yet, few techniques address this concern for Software Product Lines (SPL). In this paper, we therefore introduce two approaches to measure and reduce the energy consumption of a SPL by analyzing a limited set of products sampled from this SPL. While the first approach relies on the analysis of individual feature consumptions, the second one takes feature interactions into account to better mitigate energy consumption of resulting products.Our experimental results on a real-world SPL indicate that both approaches succeed to produce significant energy improvements on a large number of products, while consumption data was modeled from a small set of sampled products. Furthermore, we show that taking feature interactions into account leads to more products improved with higher energy savings per product. CCS CONCEPTS• Software and its engineering → Software product lines.

Configuration Optimization with Limited Functional Impact

“…It then provides optimization suggestions to the developer. ICO is inspired by the energy consumption optimization approach presented in [2], but the approach in this paper differs from the one in [2] in several aspects. In particular, the approach proposed in this paper addresses the limitations listed in [2].…”

“…ICO is inspired by the energy consumption optimization approach presented in [2], but the approach in this paper differs from the one in [2] in several aspects. In particular, the approach proposed in this paper addresses the limitations listed in [2]. That is, we propose an approach that is feature model agnostic and supports multi-objective optimization, in contrast to an optimization method that was tightly coupled to the feature model under test and dedicated to only one performance indicator, the energy consumption.…”

“…That is, we propose an approach that is feature model agnostic and supports multi-objective optimization, in contrast to an optimization method that was tightly coupled to the feature model under test and dedicated to only one performance indicator, the energy consumption. In addition, ICO supports cross-tree constraints in its optimization process, while the approach presented in [2] only focused on switching selected features based on the feature tree structure.…”

“…In a general way, each unselected feature leads to a candidate configuration where this feature is selected (lines 5-7), each selected feature leads to a candidate configuration where this feature is unselected (lines 8-10), and each exclusive relationship of both a selected and unselected features leads to a candidate configuration where the selected feature is deselected and the unselected one is selected (lines 11-13). Candidate configurations are then ordered by performance gain (line 14), and finally filtered regarding their validity and performance (line 16), to ensure that the returned suggestions (i) cannot turn a valid configuration into an invalid one and (ii) can only improve the performance of the configuration, according to the performance model 3 .…”

“…Based on this observation and inspired by our previous work on measuring the energy consumption of configurable systems [2], we propose in this paper an approach that optimizes a configuration regarding multiple performance objectives. Contrarily to prior work that samples or predicts performance models seeking for the best configuration of the whole configuration space [4,7,9,10,22], our approach optimizes existing configurations by maximizing performance gains while minimizing changes to such configurations.…”