[Paper Review] A Synthesis of Green Architectural Tactics for ML-Enabled Systems

Heli Järvenpää, Patricia Lago+4 more|arXiv (Cornell University)|Dec 15, 2023

Green IT and Sustainability4 citations

TL;DR

This paper presents a catalog of 30 green architectural tactics for machine learning (ML)-enabled systems, derived from a synthesis of 51 peer-reviewed Green AI studies and validated with software architecture experts. The tactics—organized into six categories—offer actionable, high-level design strategies to reduce energy and carbon footprints in ML system development, promoting environmental sustainability without compromising model accuracy.

ABSTRACT

The rapid adoption of artificial intelligence (AI) and machine learning (ML) has generated growing interest in understanding their environmental impact and the challenges associated with designing environmentally friendly ML-enabled systems. While Green AI research, i.e., research that tries to minimize the energy footprint of AI, is receiving increasing attention, very few concrete guidelines are available on how ML-enabled systems can be designed to be more environmentally sustainable. In this paper, we provide a catalog of 30 green architectural tactics for ML-enabled systems to fill this gap. An architectural tactic is a high-level design technique to improve software quality, in our case environmental sustainability. We derived the tactics from the analysis of 51 peer-reviewed publications that primarily explore Green AI, and validated them using a focus group approach with three experts. The 30 tactics we identified are aimed to serve as an initial reference guide for further exploration into Green AI from a software engineering perspective, and assist in designing sustainable ML-enabled systems. To enhance transparency and facilitate their widespread use and extension, we make the tactics available online in easily consumable formats. Wide-spread adoption of these tactics has the potential to substantially reduce the societal impact of ML-enabled systems regarding their energy and carbon footprint.

Motivation & Objective

• Address the lack of concrete, actionable guidelines for designing environmentally sustainable ML-enabled systems from a software architecture perspective.
• Identify and synthesize green architectural tactics that target energy efficiency in ML system development, moving beyond model-level optimizations.
• Provide a transparent, reusable, and extensible catalog of tactics to support sustainable software engineering practices in ML systems.
• Bridge the gap between Green AI research and software architecture by translating environmental concerns into design-level decisions.
• Facilitate industry adoption and further research by making the tactics publicly available in machine-readable formats and integrated into existing software engineering knowledge repositories.

Proposed method

• Conducted a systematic literature review of 51 peer-reviewed publications focused on Green AI and environmental sustainability in ML.
• Extracted and synthesized high-level design techniques (architectural tactics) from the literature that contribute to energy efficiency in ML systems.
• Organized the 30 identified tactics into six thematic categories: data-centric, algorithm design, model optimization, model training, deployment, and management.
• Validated and refined the initial tactic collection through a focus group with three expert practitioners in software architecture for ML systems (SA4ML).
• Integrated the final catalog into the Archive of Awesome and Dark Tactics (AADT) for discoverability and reuse.
• Published the complete set of tactics with provenance and metadata on Zenodo for long-term accessibility and version control.

Experimental results

Research questions

• RQ1 RQ1: Which green architectural tactics for ML-enabled systems can be synthesized from scientific literature?
• RQ2 RQ2: How do SA4ML experts perceive the synthesized collection of green architectural tactics?
• RQ3 RQ3: How can green architectural tactics be structured and formalized to support practical adoption in ML system design?
• RQ4 RQ4: What are the key trade-offs between energy efficiency and other quality attributes (e.g., accuracy) in ML system architecture?
• RQ5 RQ5: How can architectural tactics be generalized and extended to support diverse ML algorithms and system concerns?

Key findings

• The study successfully identified and synthesized 30 green architectural tactics for ML-enabled systems, organized into six coherent categories: data-centric, algorithm design, model optimization, model training, deployment, and management.
• The focus group validation confirmed the relevance and practicality of the tactics, with experts highlighting their value for guiding sustainable system design, especially in early architectural decision-making.
• Most tactics focus on model-level efficiency rather than holistic system architecture, indicating a current research gap in architectural knowledge for full-stack ML systems.
• The majority of tactics were found to maintain or minimally impact model accuracy, suggesting that significant energy savings are achievable without sacrificing performance.
• The integration of the catalog into the AADT and Zenodo ensures long-term accessibility, versioning, and reusability for both researchers and practitioners.
• The study demonstrates that architectural tactics can serve as a bridge between Green AI research and software engineering practice, enabling systematic, sustainable design of ML systems.