astroCAMP: A Community Benchmark and Co-Design Framework for Sustainable SKA-Scale Radio Imaging

2025-12-15

AI Summary

Radio interferometric imaging suffers from severe memory and I/O bottlenecks, resulting in low hardware utilization (4–14%), poor energy efficiency, high carbon emissions, and—critically—a lack of unified quality–efficiency evaluation metrics, hindering co-design of software and hardware. Addressing SKA’s sustainability requirements, this project proposes the first multi-objective co-design framework integrating scientific fidelity, performance, energy consumption, carbon footprint, and total cost of ownership. We construct a representative SKA benchmark dataset, an extensible evaluation metric suite, and a methodology for exploring heterogeneous hardware (CPU/GPU/FPGA) design spaces. Empirically validated on AMD EPYC/NVIDIA H100 platforms using WSClean and IDG, our framework incorporates lifecycle economic modeling and multi-objective optimization. Key outcomes include Pareto-optimal, quantitative quality–efficiency trade-off analysis for SKA-scale imaging; open-sourced datasets, benchmark results, and a fully reproducible toolchain; and community-driven development of fidelity-tolerance standards.

Technology Category

• Machine Learning: Hardware-aware ML
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Search and Optimization: Distributed Search

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Sustainability and carbon-aware systems for Web, mobile, and WoT
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Security and Privacy: Large-scale security measurements

Abstract

The Square Kilometre Array (SKA) project will operate one of the world's largest continuous scientific data systems, sustaining petascale imaging under strict power caps. Yet, current radio-interferometric pipelines utilize only a small fraction of hardware peak performance, typically 4-14%, due to memory and I/O bottlenecks, resulting in poor energy efficiency and high operational and carbon costs. Progress is further limited by the absence of standardised metrics and fidelity tolerances, preventing principled hardware-software co-design and rigorous exploration of quality-efficiency trade-offs. We introduce astroCAMP, a framework for guiding the co-design of next-generation imaging pipelines and sustainable HPC architectures that maximise scientific return within SKA's operational and environmental limits. astroCAMP provides: (1) a unified, extensible metric suite covering scientific fidelity, computational performance, sustainability, and lifecycle economics; (2) standardised SKA-representative datasets and reference outputs enabling reproducible benchmarking across CPUs, GPUs, and emerging accelerators; and (3) a multi-objective co-design formulation linking scientific-quality constraints to time-, energy-, carbon-to-solution, and total cost of ownership. We release datasets, benchmarking results, and a reproducibility kit, and evaluate co-design metrics for WSClean and IDG on an AMD EPYC 9334 processor and an NVIDIA H100 GPU. Further, we illustrate the use of astroCAMP for heterogeneous CPU-FPGA design-space exploration, and its potential to facilitate the identification of Pareto-optimal operating points for SKA-scale imaging deployments. Last, we make a call to the SKA community to define quantifiable fidelity metrics and thresholds to accelerate principled optimisation for SKA-scale imaging.

Problem

• Addresses low hardware utilization in radio imaging due to memory and I/O bottlenecks
• Lacks standardized metrics for evaluating scientific fidelity and efficiency trade-offs
• Needs co-design framework for sustainable HPC within SKA's power and environmental limits

Innovation

• Provides unified metrics for scientific fidelity and sustainability
• Standardizes datasets for reproducible benchmarking across hardware
• Links quality constraints to energy and cost for co-design