[Paper Review] HL-LHC Computing Review: Common Tools and Community Software

Aarrestad, Thea, Amoroso, Simone+106 more|arXiv (Cornell University)|Jan 1, 2020

Particle physics theoretical and experimental studies96 references11 citations

TL;DR

This paper reviews the state of common and community software for the High-Luminosity LHC (HL-LHC), identifying key challenges and R&D priorities across physics event generators, detector simulation, reconstruction/software triggers, and data analysis. It advocates for GPU and accelerator adoption, improved profiling and optimization, machine learning integration, and long-term sustainability of critical software through coordinated funding and career development.

ABSTRACT

Common and community software packages, such as ROOT, Geant4 and event generators have been a key part of the LHC's success so far and continued development and optimisation will be critical in the future. The challenges are driven by an ambitious physics programme, notably the LHC accelerator upgrade to high-luminosity, HL-LHC, and the corresponding detector upgrades of ATLAS and CMS. In this document we address the issues for software that is used in multiple experiments (usually even more widely than ATLAS and CMS) and maintained by teams of developers who are either not linked to a particular experiment or who contribute to common software within the context of their experiment activity. We also give space to general considerations for future software and projects that tackle upcoming challenges, no matter who writes it, which is an area where community convergence on best practice is extremely useful.

Motivation & Objective

• Address the growing computational demands of the HL-LHC, where trigger rates and pile-up will increase by ~10x and 200, respectively.
• Identify and prioritize R&D needs for shared software tools used across multiple HEP experiments.
• Ensure long-term sustainability of critical community software through funding, career development, and infrastructure support.
• Promote interoperability and portability across heterogeneous architectures (CPUs, GPUs, FPGAs, TPUs) to future-proof HEP software.
• Encourage collaboration between HEP and industry to leverage advances in machine learning and hardware acceleration.

Proposed method

• Conduct performance profiling and CPU cost accounting for physics event generators to identify optimization targets.
• Explore GPU porting via data-parallel paradigms and prototype GPU implementations for event generators.
• Advance fast simulation techniques, including ML-based and parametric models, with standardized tuning and validation frameworks.
• Develop hybrid CPU-GPU integration prototypes for Geant4 and other simulation tools, especially for calorimetry.
• Design declarative analysis description languages and metadata schemes to streamline cross-experimental analysis workflows.
• Propose interoperability schemes between HEP software stacks and mainstream data science and ML frameworks (e.g., PyTorch, TensorFlow).

Experimental results

Research questions

• RQ1 How can physics event generators be optimized for GPU and vectorized execution to reduce CPU costs?
• RQ2 What are the key performance bottlenecks in current detector simulation (e.g., Geant4), and how can they be mitigated through refactoring and specialized HEP models?
• RQ3 How can reconstruction and software trigger pipelines be made more efficient and consistent across online and offline systems under extreme pile-up?
• RQ4 What data reduction and storage models can achieve O(1 kb/event) footprints while preserving analysis flexibility?
• RQ5 How can HEP software achieve sustainable integration with external ML and data science tools without sacrificing portability or long-term maintainability?

Key findings

• GPU acceleration is identified as a top priority across all domains, especially for event generators, detector simulation (particularly calorimetry), and reconstruction.
• Current CPU performance improvements are projected at only ~10% per year, making GPU and accelerator adoption essential for HL-LHC readiness.
• Machine learning techniques show strong potential for optimizing phase space sampling, reducing negative weight events, and accelerating fast simulation.
• A major challenge lies in ensuring long-term sustainability of community software, requiring dedicated funding and career pathways for developers.
• Cross-experimental interoperability in analysis workflows can be improved through declarative syntax and standardized metadata and calibration schemes.
• Integration of ML models into C++-based HEP frameworks is seen as a key enabler for real-time analysis and trigger optimization.