The World is Not Enough: Growing Waste in HPC-enabled Academic Practice

Research Background and Issues

• Issues and Challenges:
  This paper explores the environmental impact of high-performance computing (HPC) on academic research practices. Although HPC has significantly advanced scientific research, its environmental costs, including energy consumption and waste generation, are substantial. There is currently a lack of comprehensive understanding of the growth and waste associated with HPC, such as inefficient resource utilization, hardware redundancy, and the academic culture's demand for repetitive research.

• Significance:
  Globally, the carbon footprint of ICT (Information and Communication Technology) is significant, with HPC and large data centers consuming vast amounts of energy (e.g., U.S. data centers consumed nearly 460 TWh in 2022). Projections indicate that this demand will continue to rise, exacerbating resource and environmental issues. Against the backdrop of global efforts to combat climate change and achieve carbon neutrality, the sustainability of scientific research processes urgently needs to be examined.

• Research Motivation and Related Work:
  This study is grounded in Sustainable Human-Computer Interaction (SHCI) theory, focusing on how the long-term integration of technology shapes academic practices. A literature review reveals that the issue of computational waste in HPC and academic research has not received sufficient attention. This study aims to uncover the specific drivers in this domain and provide a foundation for future interventions.

Solutions

• Proposed Approach:
  The authors conducted interviews with researchers, HPC providers, funders, and administrators to uncover the technological and cultural factors driving HPC-induced computational growth and waste. They employed Causal Loop Diagram (CLD) modeling to illustrate these complex interconnections.

• Innovative Contributions:
  The study introduces a conceptual framework for "computational waste," encompassing issues ranging from inefficient code to redundant research driven by academic culture. Through CLD, it systematically demonstrates the technical and socio-cultural factors influencing HPC growth and waste.

• Implementation Steps:

  1. Collect qualitative interview data from 25 participants in various roles.
  2. Extract and analyze key themes, such as the demand for HPC growth, forms of waste, and socio-cultural influences.
  3. Construct Causal Loop Diagrams (CLD) to link the dynamic relationships between computational growth and waste.
  4. Identify potential intervention points to address waste issues, analyzing them from technical to cultural perspectives.

Research Findings

• Key Findings:

  1. Analysis of Growth and Waste Drivers: The study identifies that computational growth often stems from high expectations and competitive pressures in academic culture, while waste includes inefficient code, redundant work, and excessive computation.
  2. Development of Causal Loop Diagrams (CLD): The diagrams depict the complex dynamic relationships between HPC integration and academic practices, revealing reinforcing feedback mechanisms between behaviors (e.g., student expectations, academic publishing culture) and facility usage.
  3. Proposed Intervention Strategies: These include developing green computing tools, enhancing resource sharing, optimizing technical support (e.g., Research Software Engineers, RSEs), and promoting sustainable computing through cultural changes, such as reducing the "publish or perish" pressure.

• Comparison with Existing Solutions:
  Compared to previous studies that focus more on technical efficiency, this research emphasizes the intersection of technical and socio-cultural factors, highlighting the importance of cultural and structural transformations in mitigating HPC waste.

• Experimental and Evaluation Results:
  Although the study primarily employs qualitative analysis, it reveals the multi-dimensional waste phenomena in academic practices through interviews and validates the relevance of the CLD.

• Limitations and Future Directions:

  1. The sample is biased toward STEM (Science, Technology, Engineering, Mathematics) disciplines, excluding social sciences and humanities. Future research should explore the unique dynamics of other disciplines.
  2. The study relies on qualitative methods, necessitating quantitative data to assess the scale and impact of waste practices.
  3. Current solutions focus more on technical improvements, while detailed implementation of policy optimization or interdisciplinary cultural shifts requires further investigation.

Conclusion

Through an in-depth analysis of HPC growth and waste, this paper elucidates the sustainability challenges in academic practices, emphasizing the importance of addressing the interplay between academic culture, structure, and technology. The research not only provides a new perspective on environmental issues in the digital transformation but also proposes insightful intervention strategies to guide scientific research toward more responsible and meaningful directions.