Tools and best practices for sustainable software.pdf
- 1. GREEN CODING; Tools and best practices for Sustainable Software An overview
- 2. GREEN CODING; Quick info - Arne Tarara / Green Coding Berlin 2 Standard slide for starters • Software-Dev 16+ years • Founder & CEO Green Coding Berlin GmbH • We do research and development in open source tools for software resource consumption • We help companies assess their digital infrastructure resource consumption
- 3. GREEN CODING; 3 Agenda • What is sustainable software engineering? • Which tools and concepts can I use? • What are best practices? • Why is sustainable software hard? 25.04.2023
- 4. GREEN CODING; 4 What is the definition of sustainability ... in the domain of hard- and software Credit: SDIA
- 5. GREEN CODING; To make software more sustainable we need to make its consumption visible. 5 Software uses energy and machines
- 6. GREEN CODING; To make software more sustainable we need to make its consumption visible. ... so what tools do already exist? 6 Software uses energy and machines
- 7. GREEN CODING; open-source RAPL based command line tool 7 Easy starter: Scaphandre - Hubblo • Neat feature: Can split by process
- 8. GREEN CODING; codecarbon.io Ready to use tools • Python • RAPL-based • NVIDIA GPU support
- 9. GREEN CODING; Introducing Kepler 9 Distributed Environments / Clusters http://sustainable-computing.io
- 10. GREEN CODING; All-in-One solution for benchmarking, orchestration and transparency 10 Green Metrics Tool
- 11. GREEN CODING; 11 Green Metrics Tool Use-Case mit Wordpress / Django Community Why is there server and database activity here? Container native; For detailed inspection of architecture
- 12. GREEN CODING; 12 Green Metrics Tool Statistical comparisons and API included
- 13. GREEN CODING; Energy inside of a CI / CD Pipeline 13 Eco CI Tools - Green Coding Berlin Eco-CI Tools on Github - Integrated into Actions directly
- 14. GREEN CODING; 14 Weitere Tools ... • powertop • powermetrics • Cloud Carbon Footprint • turbostat • powerJoularX • PAPI • greenframe.io • ... for questions regarding a speci fi c tool, please ask in the Q&A!
- 15. GREEN CODING; Which should I use? And how do I interpret the data? Are best practices for sustainable software? 15 Ok, so know I know some tools
- 16. GREEN CODING; 16 Best Practices • Measure your system • Go for reusability • Go for exportability / interoperability • Code must also run on older hardware • To fi ght device obsolescence (within reason) According to the Blue Angel for Software
- 17. GREEN CODING; 17 Best Practices • Tips that are well known and not really "green software" • Cache Static data • Use energy e ffi cient AI models 🤡 • Minify JS / CSS • Right-Size VMs • ... • Tips that are debatable • Run AI models at the edge • Use sustainable regions -> Depends on the network transfer. Hard to wage • Stateless design -> On its own no gain • Source: https://patterns.greensoftware.foundation/catalog/cloud/match-utilization-requirements-of-vm/ According to the Green Software Foundation - A suprising mix ....
- 18. GREEN CODING; 18 Best Practices? • Is email more sustainable than paper? • Paper consumes a fi xed amount. Email has pot. in fi nite storage and processing • Is Serverless more sustainable than classic VMs? • No solid data on this (Deno / Isolates / Firecracker) • Cloud fl are / Amazon did decline when asking for sustainability insights • Is using AWS Gravitron more sustainable than Intel • What happens to electronic waste?! Life-Cycle ...? • Is Python more sustainable than Rust? • Python uses 80-times the Instructions where as Rust uses 1-3. Still people are not changing because of "cost of development" etc. .... Even simple questions are hard (impossible) to answer
- 19. GREEN CODING; 19 Problems • Often they just employ common sense - Like: "run less tests" • Often generic advice - "Use sustainable libraries" • Often can be harmful also - Switching machines brings electronic waste • Hard to generalize at the moment - Because we have no data • Can contain pitfalls Like: "Less runtime is always better" ... no: TurboBoost, HyperThreading etc. may inverse result • We plead for a simpler solution .... with current "sustainability" best practices
- 20. GREEN CODING; 20 Best Practices • You just have to measure! • You have to ask (service providers) for the metrics you don't know. • And to treat energy and CO2 as a fi rst order metric .... constantly (git-ops)! • We need a lot of data for general statements • And we need to include the development and delivery phase of software • Isolated benchmarks and claims are helpful for a isolated view, but not for global sustainability But how do we get there
- 21. GREEN CODING; 21 Best Practices 1. Quantify your system! How much am I using for what service (Green Metrics Tool, Scaphandre, Kepler etc.) 2. Green Energy (-> Green Web Foundation) But not on the moon please! 3. Idle machines (Async / Polling / Microservices) Only makes sense if you can really turn machines o ff 4. Architecture Overhead (Backups, Logs, Redundancy, Over-Provisioning of Services / Machines) 5. Language Overhead (C-Extensions, WebAssembly, Language-Swaps) 6. Code Changes Even 99% Code runtime reduction is of limited help if the machine then idles An approach for a framework atm.
- 22. GREEN CODING; 22 Best Practices • No fancyness until proven helpful • Use less resources -> Yes, common sense! • Services, Machines etc. • No o ff -loading / Backlashes • We are more sustainable, we now use a data provider for that machine! • We now use 5 GB less memory (and calculate everything on request :( ) • General Question: If everybody would do it this way. Would we have enough resources? • Simple chat-bot => ChatGPT • Portfolio Website: VPS with Wordpress and database Some general guidance
- 23. GREEN CODING; Follow green-coding.berlin 23 Thank you! Time for Q&A • Website / Blog / Newsletter: https://www.green-coding.berlin • Demo Open Data Repository: https://metrics.green-coding.berlin • Unsere Projekte: https://www.green-coding.berlin/#projects • Unsere Case-Studies: https://www.green-coding.berlin/case-studies • Meetup Gruppe (Berlin): https://www.meetup.com/green-coding • https://www.linkedin.com/in/arne-tarara / arne@green-coding.berlin
- 26. GREEN CODING; 26 Green Metrics Tool - Schaubild Container Aufbau für Client-Server Apps
- 27. GREEN CODING; 27 Green Metrics Tool - Hardware Setup • AC / DC Energie • IPMI • GUDE (Blauer Engel) • PowerSpy2 - Consumer Grade Oscilloscope • Custom Mainboard Connector for Fujitsu ESPRIMO • CPU % / CPU-Frequenz • Temperatur • CPU / DRAM Energie • RAPL • Netzwerk Datenverkehr / Energie • Container-Metriken • ... weitere Reporter als Open-Source Plugin-In möglich Mess-Sensoren im Green Metrics Tool
- 28. GREEN CODING; 28 Green Metrics Tool - Cluster Setup • Fujitsu ESPRIMO P956 - Blue Angel compatible (Ubuntu) • Fujitsu TX1330 M2 - Single-Tenant Server (Ubuntu) • Quanta Leopard - Multi-Tenant Server - SoftAWERE compatible (Ubuntu) • Intel Mac 13" Q3-2015 • M1 Mac 13" Q1-2022 Aktuelle Rechner im Green Metrics Tool Cluster
- 29. GREEN CODING; Software Kategorien und OS 29 Green Metrics Tool - Supported Software • Server / Client Anwendungen • Beispiele Nextcloud und Django CMS • CLI Anwendungen • Beispiele curl oder ML Anwendungen (sklearn, pytorch) • Desktop Apps • Beispiele: Google Chrome, Firefox etc. • OS • macOS • Linux (Ubuntu & Fedora) • Windows (nur Gesamtleistung an der Steckdose)