Green IT Practices

Sources

§Section titled Sources

• Green Software Patterns
• Green Software Foundation
  • Articles
  • Book recommendations
• 10 Recommendations for Green Software Development
  • Papers
    • IEEE 2015: Understanding Green Software Development: A Conceptual Framework
      • Modelling and profiling
      • 1. Refactoring (energy smells (see paper below)) and 2) self-adaptation.
      • Includes guidelines.
  • Greening of IT vs. Greening by IT.
  • Red AI: “throwing more computing power at a problem to get better results than necessary”
    • Learn to draw the line in terms of quality requirements.
    • More discussion about Green AI vs Red AI here.
• Definition, Implementation and Validation of Energy Code Smells: an Exploratory Study on an Embedded System
  • The smells seem to be only about dead or redundant code.
  • Effect seems extremely minor (micro watts).

Approaches

§Section titled Approaches

• See the Green Software Patterns catalog. The items below are a collection of others I found.
• Good software architecture
  • Focus on using good higher-level abstractions.
• Use an appropriate data structure / code
  • IDE based recommendations for use of efficient data structure implementations (correctness is respected of course; Java).
  • Algorithms that are most energy efficient on the common case (scenario-driven refactoring).
• Use the appropriate programming language (discussion)
• Optimise hot code (Amdahl’s law basically; can use energy profiling)
• Clean up useless code and data
• Execute less
  • Selective testing as discussed in Testing.
  • CI/CD pipelines running less code.
• Analytical (?) over AI? (software 1.0 over software 2.0)
  • Presuming that specialized handwritten code is more energy efficient than AI models (also consider the training).
• Good enough is good enough
  • Lower the accuracy of measurements
  • Store less / only as much as you really need
  • Reduce number of invocations (e.g., poll a server less frequently). Also, preempt instead of busy polling?
  • See red AI below.
  • Feels similar to the use of risk budgets in SRE.
  • Discussed in Microsoft: Sustainable Software Engineering (SSE) and the role and responsibilities of a Sustainable Software Engineer
• Reduce data usage.
  • Caching (e.g., locally to avoid computations or CDN based for less traffic).
  • Compression.
  • Smaller digital media (smaller images).
• Self-adaptation: software with multiple energy profiles (discussed; e.g., low power mode turns off background processes; might require realtime energy profiling for profile selection).
• Appropriate hardware: size (phone vs. server), processor type, and modernity (newer hardware better? but tradeoff on environmental e-waste and resource use).
• Product: Minimalism; avoiding scope creep.
  • Remove unused features (also good for maintainability)
• Green robustness. Use of chaos engineering to find inefficiencies. The Carbon Monkey.