Resource-aware Research on Universe and Matter: Call-to-Action in Digital Transformation

Drawing from a May 2023 workshop, this paper calls for resource-aware research in the fields of Universe and Matter by outlining a portfolio of digital transformation measures designed to simultaneously advance scientific progress and mitigate climate change through reduced fossil fuel reliance.

The Gist

Imagine the scientific community studying the universe (like black holes and galaxies) and matter (like tiny particles) is like a massive, high-tech kitchen. For years, this kitchen has been cooking up incredible recipes (scientific discoveries) using a stove powered by fossil fuels. But the planet is overheating, and we need to switch to a clean, green stove before the house burns down.

This paper is a call to action from a group of scientists who say: "We can't just keep cooking the same way. We need to change our recipes, our tools, and our habits to save energy without stopping the cooking."

Here is a simple breakdown of their plan, using everyday analogies:

1. The Problem: The Kitchen is Running Hot

The scientists explain that climate change is real and dangerous. It's like a pressure cooker that's about to blow. To stop it, we need to cut our carbon emissions in half very quickly.

• The Issue: Doing science today requires massive amounts of computer power. These computers are like giant ovens that eat electricity. If that electricity comes from coal or gas, the scientists are accidentally contributing to the climate crisis.
• The Goal: They want to keep making amazing discoveries but do it in a way that doesn't hurt the planet.

2. The Solution: A "Smart Kitchen" Approach

The paper suggests six main ways to make their research "green." Think of these as rules for a sustainable kitchen:

A. Smart Data: Don't Hoard Everything

• The Old Way: Imagine taking a photo of every single grain of sand on a beach, storing them all, and then trying to find one specific grain later. It wastes space and energy.
• The New Way: Be a "smart filter." Only keep the photos that actually matter. If you have a photo of a grain of sand that looks exactly like one you already have, delete the copy.
• The Trick: Make sure your "photos" (data) are labeled clearly so others can find and use them later without you having to take the picture again. This is called making data FAIR (Findable, Accessible, Interoperable, Reusable).

B. Better Software: Fix the Recipe, Don't Just Cook Faster

• The Old Way: Writing messy code is like writing a recipe on a napkin that only the original chef understands. If that chef leaves, the recipe is lost, and someone else has to rewrite it from scratch, wasting time and energy.
• The New Way: Write clean, organized recipes (code) that anyone can read and use. If you find a way to cook a dish in 10 minutes instead of 20, do it! But don't just use that extra time to cook more dishes; use it to save energy.
• The Goal: Stop "reinventing the wheel." Use tools that already exist and are proven to work.

C. Smarter Algorithms: The AI Sous-Chef

• The Idea: Artificial Intelligence (AI) is like a super-fast sous-chef. It can chop vegetables (analyze data) much faster than a human.
• The Catch: Training the AI takes a lot of energy (like running a blender for hours). The scientists say: "Use the AI, but only for the big jobs where it saves the most energy in the long run." Don't use a jet engine to power a bicycle.
• The Rule: If a human can solve a problem with a simple thought, don't use a massive computer to do it.

D. Green Locations: Move the Kitchen to the Wind

• The Idea: Currently, data centers (the kitchens) are often in cities where electricity comes from dirty power plants.
• The New Plan: Move the data centers next to the source of clean energy. Imagine building a data center right next to a giant offshore wind farm. When the wind blows, the computers run. When the wind stops, the computers sleep or slow down.
• Heat Recovery: Computers get hot. Instead of letting that heat go to waste, use it to warm up nearby houses or offices. It's like using the heat from your oven to warm your kitchen in winter.

E. Longer-Lasting Tools: Fix, Don't Toss

• The Problem: We throw away computers too fast. Making a new computer creates a lot of pollution (like manufacturing a new car).
• The Fix: Keep the old computers running longer. Just because a computer isn't the "latest model" doesn't mean it can't do the job. Extend its life, just like you would repair an old jacket instead of buying a new one every year.

F. Education: Teach the Next Generation

• The Goal: We need to teach young scientists that "saving energy" is just as important as "finding new facts."
• The Mindset: Before hitting "run" on a massive calculation, ask: "Do I really need to do this? Is there a cheaper way?" It's about balancing the value of the discovery against the cost of the energy used.

3. The Timeline: What Can We Do When?

The paper organizes these ideas into three buckets:

• Do This Now (Short Term):

  • Start talking about climate change in the lab.
  • Check how much energy your specific computer jobs are using.
  • Make a plan to reduce your carbon footprint.

• Do This in a Few Years (Medium Term):

  • Organize your data so it can be reused easily.
  • Rewrite messy software to be efficient.
  • Train everyone on how to write "green" code.

• Plan for the Future (Long Term):

  • Build new data centers near wind farms.
  • Create software that automatically slows down when the wind stops blowing.
  • Design systems that capture and reuse heat.

The Bottom Line

The scientists are saying: "We love science, and we love the planet. We can't choose one over the other. By being smarter about how we use our computers, we can continue to unlock the secrets of the universe without burning down our home."

They are asking for a partnership between scientists, computer experts, and governments to make this transition happen, ensuring that the future of research is as bright as the stars they study.

Technical Summary

Technical Summary: Resource-aware Research on Universe and Matter: Call-to-Action in Digital Transformation

Problem Statement
The paper addresses the urgent conflict between the escalating computational demands of research on the Universe and Matter (ErUM) and the global imperative to reduce greenhouse gas (GHG) emissions to avoid climate tipping points. While the Paris Agreement necessitates a 50% reduction in global emissions within seven years, the digital infrastructure supporting ErUM research—characterized by massive data rates from high-energy physics, astrophysics, and photon/neutron sciences—consumes significant energy. The authors note that data centers currently account for approximately 1% of global energy consumption, a figure projected to rise as data volumes grow (e.g., from Terabits per second at future facilities like the Square Kilometre Array). The core challenge is to maintain scientific progress and knowledge gain while drastically reducing the carbon footprint of computing, storage, and algorithmic processing.

Methodology
This work is not an experimental study but a strategic synthesis based on a three-day workshop held in May 2023 involving scientists from the ErUM-Data community in Germany and international collaborators. The methodology involved:

1. Structured Deliberation: The workshop was guided by twelve pre-formulated questions (detailed in Appendix A) covering hardware, data, algorithms, energy supply, and cultural shifts.
2. Portfolio Development: Participants analyzed current practices and identified actionable measures across six specific domains: smart data transformation, software engineering, algorithms/AI, computing infrastructure, education, and institutional support.
3. Categorization by Feasibility: The identified measures were categorized based on the time and effort required for implementation: immediate/short-term, medium-term (few years), and long-term (coordinated planning).
4. Contextual Analysis: The authors integrated existing data on energy consumption (e.g., PUE values, specific GWh usage of German computing centers) and emission scopes (Scope 1, 2, and 3) to ground their recommendations in technical reality.

Key Contributions
The paper proposes a comprehensive "portfolio of measures" to transition ErUM research toward sustainability. The contributions are organized into six technical and organizational areas:

1. Smart Data Transformation:

   • Advocates for FAIR data (Findable, Accessible, Interoperable, Reusable) to maximize reuse and minimize redundant processing.
   • Proposes "stream-based models" where raw data is processed online and discarded if not scientifically valuable, rather than storing all raw data.
   • Suggests optimizing the storage of intermediate workflow results (snapshots) to balance storage costs against the energy required for re-calculation.

2. Software Engineering and Data Analysis:

   • Emphasizes the transition from "spaghetti code" to professional, modular, and version-controlled software to prevent "abandonware" and ensure reproducibility.
   • Calls for the integration of energy efficiency as a primary software quality metric, alongside standard benchmarks.
   • Recommends automated parallelization, vectorization, and adaptation to new CPU architectures (e.g., ARM, RISC-V) to improve energy-per-operation efficiency.

3. Algorithms and Artificial Intelligence (AI):

   • Highlights the potential of generative models and pre-trained models to replace computationally expensive simulations (e.g., replacing Geant4 detector simulations with deep generative models that are orders of magnitude faster).
   • Argues for the strategic use of AI only where it offers significant scientific benefit or resource savings, warning against the "rebound effect" where efficiency gains lead to increased total consumption.
   • Stresses the need to document the CO2e footprint of both training and inference phases of machine learning models.

4. Computing and Infrastructure:

   • Proposes relocating data centers to be co-located with renewable energy sources (e.g., offshore wind farms) to utilize direct power and reduce transmission losses.
   • Suggests dynamic workload scheduling where computing resources scale up or down based on real-time renewable energy availability (e.g., "malleable jobs").
   • Recommends extending hardware lifetimes (beyond the typical 5 years) where feasible to amortize the manufacturing carbon footprint (Scope 3), and reusing waste heat for district heating.
   • Calls for improved Power Usage Effectiveness (PUE) and the implementation of energy storage solutions to buffer against renewable intermittency.

5. Education and Culture:

   • Identifies a need for new curricula to train scientists in sustainable digital practices, including resource-aware coding and the ethical implications of AI.
   • Advocates for a cultural shift where researchers actively balance knowledge gain against resource usage.

6. Funding and Institutional Support:

   • Calls for funding agencies to support long-term positions for software and data management and to include carbon footprint assessments in project planning.

Results and Proposed Actions
The primary result of the paper is Table 2, a prioritized "Call-to-Action" list categorized by implementation timeline:

• Immediate/Short-term: Raising awareness, monitoring energy consumption at the job level, and considering carbon footprints in project planning.
• Medium-term (Few years): Implementing FAIR data practices, optimizing software for energy, adopting workflow management, and developing tools for resource monitoring.
• Long-term: Establishing data centers near renewable energy sources, developing dynamic middleware for energy-aware scheduling, and optimizing hardware lifecycles.

The paper also provides specific technical data, such as the breakdown of CERN's energy consumption (55% processing, 21% disk storage) and the potential for reducing node power consumption by 50% through clock rate throttling.

Significance and Claims
The authors claim that the ErUM community possesses the tools and expertise to develop a realistic plan for reducing CO2e emissions without sacrificing scientific progress. They argue that sustainability is not merely an environmental constraint but a driver for innovation in algorithms, software architecture, and infrastructure management.

The paper asserts that the transition requires a "difficult, but unavoidable" shift in attitude and behavior across all levels of the scientific community. It positions the proposed portfolio of measures as a necessary framework to align the digital transformation of ErUM research with the goals of the Paris Agreement. The authors conclude that by consciously balancing prospective knowledge gain with resource usage, the community can achieve efficiency gains that will ultimately accelerate scientific progress while preserving the planetary conditions necessary for life.