Source-linked AI summary
The Green Software Landscape: A Systematic Mapping Study on Evolution, Applications, Software Lifecycle, and Best Practices
Max Hort, Maria Kechagia, Federica Sarro
TL;DR
Software’s energy and climate impacts motivate Green Software Engineering research. This study collected and analyzed 390 publications, finding that mobile is the largest application domain and that AI-related sustainability studies increased in 2023 and 2024.
Problem
Green Software Engineering addresses software’s energy consumption and impact on global warming.
Method
The study collected and analyzed 390 publications and examined experimental-design practices including stability in energy-related studies.
Results
390 publications were analyzed; mobile was the largest application domain with 152 publications, while AI sustainability applications increased in 2023 and 2024.
Takeaways & Limitations
The study aims to inform the Software Engineering community about Green Software approaches and support dissemination of this research.
Takeaways & Limitations
The reviewed literature may be incomplete because relevant publications could have been missed.
Abstract
from arXiv · showhide
Energy consumption and climate change have made sustainability critical in Software Engineering (SE), driving the emergence of Green SE. Over the past 15 years, numerous solutions for sustainable software systems have been published by the SE community, offering a rich resource for analyzing the field's evolution. To explore this, we conducted a systematic mapping study of Green SE research published between 2010 and 2024. We collected 390 publications, categorizing them by application domain (e.g., mobile, cloud, AI) and research type (e.g., optimisation study, benchmarking, literature review Additionally, we analyzed a representative subset of 79 papers to classify the key elements-such as hardware, measurement, stability, and replicability-considered during energy measurement experiments. Our findings indicate that SE conferences host the majority of energy-related literature. Notably, Green SE studies surged in popularity starting in 2023, largely driven by AI-related publications. Optimization and benchmarking emerged as the most prevalent research types. Ultimately, we aim to inform the SE community about current approaches to energy concerns, highlight critical experimental practices, and advocate for continued action toward more sustainable software engineering.
1 Introduction
Green Software Engineering has grown rapidly, but its energy-related research remains fragmented across venues and application targets. This study maps 390 publications from 2010–2024 across venues, applications, research types, and experimental practices.
- Motivation: The field is steadily increasing but remains fragmented across journals, workshops, and conferences with different goals and application targets.The study aims to inform the software engineering community about current approaches and encourage further Green SE action.
- Study scope and aims: 390 publications from 2010–2024 are classified across applications, methods, and experiments to provide a comprehensive view of Green SE.The study examines top software engineering venues and categorizes applications, study types, and experimental-design elements.
- Publication venues: 183 of 390 publications (47%) appear in top software engineering conferences, compared with 34% in workshops and 19% in journals.Conferences therefore host the largest share of energy-related Green SE literature.
- Application domains: Mobile computing is the most frequently investigated application domain, while artificial intelligence is a fast-growing area for energy research.The study identifies application domains including mobile, artificial intelligence, and cloud computing.
- Research types: Optimisation and benchmarking are the most prevalent Green SE research types.The study distinguishes research types such as optimisation and benchmarking across the collected publications.
- Experimental practices: Energy-measurement experiments most critically consider hardware requirements, measurement tools, stability, replication, and benchmarks.These elements are examined as part of the experimental design protocols in the mapped studies.
2 Background
Green SE addresses software energy consumption within broader software sustainability, which concerns environmental, social, and economic impacts across the development life cycle. A systematic mapping study organizes this research by classification and publication frequencies across software engineering knowledge areas.
- Green SE and sustainability: Green SE focuses on reducing energy consumption across software engineering tasks and processes.The survey uses Green SE for work addressing software energy concerns.
- Green SE and sustainability: Sustainable software is commonly described as minimizing negative societal, economic, and environmental impacts throughout the software development life cycle.The survey emphasizes continuous reduction of resource consumption, including energy usage.
- Systematic mapping studies: A software engineering systematic map builds a classification and organizes a field according to that classification, emphasizing publication frequencies across categories.This approach differs from systematic reviews, which aggregate evidence to answer specific questions.
- Systematic mapping studies: The study chooses systematic mapping to investigate Green SE evolution over 15 years, including the emerging Green AI subfield.Mapping studies use broad questions to reveal research trends such as publication evolution and covered topics.
- Knowledge-area framework: The classification uses four SWEBOK-aligned knowledge areas: requirements, architecture, construction, and testing.Together, these areas cover core development-life-cycle phases and cross-cutting software engineering concerns.
- Knowledge-area framework: Green SE studies examine energy concerns in requirements, architecture, implementation, optimization, and testing measurement.The categories respectively address energy requirements, energy-influencing architecture, implementation choices, optimization, and testing activities.
3 Study Design
The study uses a systematic mapping design to classify Green Software Engineering research and examine how energy-related experiments are designed. Four research questions cover field evolution, application domains, research approaches, and experimental protocols.
- Study design: The study provides an overview of Green Software Engineering through publication classification and frequency metrics.Its protocol defines the research questions and procedures used to achieve the study goals.
- Research questions: RQ1 examines sustainability-related publications in top software engineering venues from 2010 to 2024, including venue type and publication volume.Two authors cross-check the publications and categorize the collected data.
- Research questions: RQ2 identifies application domains addressing energy concerns and highlights fields requiring further green-aware development.The analysis uses application-field categories such as mobile applications and the ACM Computing Classification System.
- Research questions: RQ3 summarizes research types and tasks used to reduce energy consumption across software engineering application fields.Two authors manually classify research types and tasks using IEEE’s Software Engineering Body of Knowledge classification.
- Research questions: RQ4 analyzes 79 empirical studies to identify experimental design elements supporting reliable and reproducible Green Software Engineering research.The elements are grouped into hardware, stability, repetitions, measurement, and datasets.
3.2 Search Strategy
The search strategy manually targets prominent software engineering venues over a 15-year period, supplemented by co-located events and established digital libraries. This approach is intended to capture influential and emerging Green Software Engineering research.
- Search rationale: The study focuses on top-tier venues to capture field trends and surface research directions relevant to researchers and practitioners.The authors associate high-quality publications with advanced research and broader community impact.
- Search inventory: Table 1 lists the venues considered in the search procedure.The table is complemented by Table 2, which lists co-located events with Green Software Engineering publications.
- Venue selection: The search covers conferences and journals ranked A or A* by CORE, along with their co-located events.Co-located workshops can reflect emerging trends and specialized Green Software Engineering topics.
- Publication retrieval: The researchers manually inspect official venue websites and consult ACM, IEEE Xplore, or DBLP when papers cannot be retrieved directly.The procedure prioritizes direct retrieval from selected conferences, journals, and associated events.
- Search period: The search spans 2010–2024 to provide a 15-year overview of Green Software Engineering research.Limited availability and maintenance of earlier conference and workshop websites motivated 2010 as the starting year.
3.3 Selection Criteria
Selection criteria require studies to address both software engineering and energy-related sustainability. Additional RQ4 criteria narrow the experimental-design analysis to sufficiently detailed optimization or benchmarking studies.
- General criteria: Studies must address software engineering and sustainability, specifically energy-related concerns.These inclusion criteria are derived from the study’s research questions.
- General criteria: 132 publications are filtered out, leaving 390 publications for the systematic mapping study.The criteria are intended to retain relevant primary studies.
- RQ4 criteria: RQ4 includes optimization studies that reduce energy consumption or benchmarking studies comparing implementation choices by energy use.The benchmarking example compares Java collection types.
- RQ4 criteria: RQ4 excludes publications with six or fewer pages to ensure space for comprehensive empirical results and design choices.The criterion targets reporting completeness rather than topic relevance.
- RQ4 criteria: Hardware- and cloud-based studies, plus approaches optimizing testing energy, are excluded from the RQ4 analysis.The analysis focuses on software systems’ core components executed by users.
- RQ4 criteria: 79 publications remain for extracting experimental design protocols.The RQ4 subset is selected from the 390 publications investigated.
3.4 Selection Process
The selection process begins with manual venue searches and title screening, followed by abstract and full-text review when necessary. It produces a final dataset of 390 primary studies from an initial pool of 521 potentially relevant records.
- Relevance verification: The researchers manually verify sustainability-related terms in included titles, including “green,” “energy,” “sustainability,” and “power.”This verification follows title-based selection of primary studies.
- Initial screening: Manual screening across selected venues and 33 co-located events identifies studies whose titles suggest sustainability relevance.The initial title screening covers conferences, journals, and co-located events.
- Relevance verification: Abstracts and full texts are reviewed when necessary to determine whether papers address sustainability and energy-related concerns.This review excludes studies that remain irrelevant after title screening.
- Final dataset: 132 studies are excluded during the selection process, yielding a final dataset of 390 primary studies.The final dataset’s annual distribution is presented in Figure 1a.
- RQ4 subset: The RQ4 filtering is deterministic because it uses categories already assigned for RQ2 and RQ3.The authors consider one validation adequate for this additional filtering.
3.5 Data Extraction
The study recorded publication metadata for RQ1–RQ3 and, for RQ4, extracted experimental-design information from a filtered subset to support reproducibility.
- Publication metadata: Each primary study was coded by source, title, year, application domain, and study type.Study types included benchmarking, empirical study, and software-engineering tasks.
- Experimental practices: The RQ4 extraction recorded hardware, energy-measurement approach, replication, stability handling, and datasets.Stability included treatment of noise and overheads such as warmup and cooldown.
- Reproducibility: The study made results from each search and classification stage publicly available to support reproducibility.The reported extraction and classification process was intended to be inspectable.
3.6 Manual Labelling
Two authors manually labelled publications by application domain and research type, resolving disagreements through discussion and additional paper-text review when needed.
- Application domains: Two authors independently examined publication titles to assign application domains such as mobile, cloud, and AI.Disagreements were discussed and, when necessary, checked against abstracts or full texts until consensus.
- Application domains: 81% agreement was achieved for application-domain labels.The reported agreement concerns the RQ2 labelling process.
- Research types: 75% agreement was achieved for subcategories describing methods or approaches reported in the publications.The same general independent-review and disagreement-resolution procedure was applied.
3.7 Threats to Validity
The mapping study faces limitations in literature completeness, manual categorisation, RQ4 filtering, and venue coverage, although it reports a broad 390-publication corpus and reproducibility materials.
- Internal validity: Publication completeness cannot be guaranteed, despite reviewing all selected top SE venues and applying inclusion and exclusion criteria.The authors present the venue review and criteria as mitigation rather than elimination of the threat.
- External validity: Relevant publications may have been missed because the search did not cover every SE venue or use keyword-based searching.The authors specifically identify possible omissions from venues such as performance engineering and sustainable-society outlets.
- External validity: ICT4S publications were excluded because the venue did not meet the study’s A/A* SE-venue criterion.The authors propose examining that venue’s applications and comparing its findings with this survey in future work.
- Scope: The survey collected 390 publications and describes the corpus as highly representative despite its scope limitations.The stated purpose was to provide a comprehensive overview of Green SE.
- Internal validity: Manual categorisation of application fields and study types may contain errors, although two authors independently labelled studies and resolved disagreements.The authors provide labelling decisions in an online repository to make this step reproducible.
- Internal validity: The deterministic RQ4 filter may exclude additional papers that would enter the 79-paper subset under more lenient criteria.The authors note that broader categories or fewer-page requirements could produce more eligible papers.
4 Results
Green SE research is broadly disseminated but unevenly concentrated across venues and has shifted from hardware and mobile toward AI. Development-focused optimisation and benchmarking dominate, while experimental practices vary across tools, devices, and protocols.
- Publication trends: Publication activity peaked in 2016, declined through 2022, and rose again in 2023–2024, which the authors conjecture is linked to growing interest in greener AI.
- Publication venues: 183 of 390 publications appeared in conferences, compared with 133 in workshops and 74 in journals.
- Publication venues: JSS had the highest publication count among 50 venues and was the only journal with more than ten relevant publications.
- Publication venues: 24 of 50 venues contained only one relevant publication, indicating broad dissemination alongside limited thematic concentration in many venues.
- Application domains: Mobile was the largest application domain with 152 publications, followed by SE with 119, Hardware with 81, AI with 32, and Others with six.
- Application domains: AI activity was near zero until 2020, grew sharply from 2021, and became the leading domain by 2024, while Hardware declined after its 2013 peak of approximately 13 papers.
- Study types and lifecycle: Development was the most studied lifecycle area, with 119 optimisation and 88 benchmarking publications; Architecture and Requirements had only four and 17 publications, respectively.
- Experimental practices: Energy studies used 31 distinct measurement or estimation tools, while execution-environment control and repetition protocols reflected a trade-off between reliability and feasibility.
5 Implications
The authors argue that Green SE should receive stronger institutional support across venues, education, tooling, collaboration, and experimental practice. They emphasize making energy measurement and reproducibility routine while expanding real-world evidence.
- Venues: SE venues could require reviewers and editors to inquire systematically about energy measurements, alongside dedicated events, invited talks, and data challenge tracks.
- Education: Education initiatives such as seminars, tutorials, summer schools, and advanced schools could build researchers’ sustainability and energy-measurement competencies.
- Research–practice synergy: Progress requires large-scale real-world datasets, field experiments, and systematic collaboration between academics and practitioners.
- Reproducibility: Clear guidelines, standardised benchmarks, open science, and containers such as Docker could make Green SE experiments more systematic and reproducible.
- Practical boundary: Expensive hardware and real-world data constrain some experiments, making research funding and industry collaboration practically relevant.
- Application priorities: The mapping indicates that mobile has received the most attention, whereas AI for SE has only begun exploring energy concerns despite its rapid growth.
- Tooling: Easy-to-use, reliable, scalable, and low-cost tools should measure energy consumption and report results to users to increase energy awareness.
6 Related Work
Earlier Green SE surveys covered smaller or earlier corpora, while this mapping study examines Green SE through 2024 across trends, domains, lifecycle phases, and empirical practices. Its comparison highlights a shift toward AI and development-focused research, with architecture and requirements still less explored.
- Survey scope: Earlier mapping studies covered 79–83 studies up to 2013 or 2017, while a later survey collected 101 papers from 2014 to 2021.
- This study: This study maps Green SE publications from 2010 to 2024 across publication trends, application domains, software-development lifecycle phases, and empirical practices.
- Publication trends: Prior surveys generally found conferences to be the predominant venue, while this study additionally identifies JSS as a major Green SE journal venue.
- Application domains: Mobile remains the largest application category, followed by AI, while Green AI surveys separately examine energy efficiency across architecture, training, inference, and data usage.
- Lifecycle coverage: Development or construction is the most extensively studied lifecycle area, whereas Architecture and Requirements remain comparatively underexplored.
- Research focus: Earlier studies emphasized sustainability as a software requirement, whereas post-2020 work increasingly assumes it and focuses on methods, techniques, and tools for construction.
- Empirical practices: The study groups key experimental elements into hardware, measurement, stability, replication, and datasets, and makes survey results publicly available to facilitate replication where possible.
7 Conclusion
This systematic mapping study analyzed 390 Green SE publications and a subset of 79 papers to characterize application domains, research types, and empirical energy-measurement practices. Mobile applications and optimization or benchmarking studies were prominent, while recent AI-focused growth and gaps in experimental reporting and datasets point to continuing research needs.
- Applications: 152 publications addressed mobile applications, making mobile the largest application domain examined.
- Applications: AI sustainability applications, including deep neural networks and large language models, increased in popularity during 2023 and 2024.
- Research types: 53% of the 390 publications proposed optimization approaches or conducted benchmarking to recommend energy-friendly implementations.
- Experiments: The 79-paper empirical analysis identified 31 measurement tools and emphasized repeated runs, warmup or cooldown steps, and systematic reporting of study settings.These practices address experimental stability, noise, and replicability.
- Future directions: The study calls for large-scale real-world datasets and empirical studies to evaluate the sustainability of modern software systems.The authors also make labelled search results and extracted information publicly available for reproducibility.