COMPASS: Community Petascale Project for Accelerator Science and Simulation
The focus of TASCS-COMPASS collaboration is the development of high-performance components for accelerator modeling, with initial emphasis on particle tracking, beam optics, and solvers for space charge models. Also, the selection and parameterization of accelerator components motivates TASCS work on computational quality of service (CQoS), a TASCS component technology initiative.
Collaboration Status: Active
Partner Project Contact: Jim Amundson, Fermilab amundson@fanl.gov
TASCS Contact: Lois Curfman McInnes, ANL, curfman@mcs.anl.gov
Collaboration Summary: A key concept in the COMPASS project, which focuses on developing a comprehensive set of interoperable components for end-to-end multi-physics simulations involving beam dynamics, electromagnetics, electron cooling, and advanced accelerator modeling, is the notion of software infrastructure for multiphysics-based accelerator modeling on petascale architectures. This effort demands a high degree of team collaboration in order to manage the integration of a wide range of accelerator physics applications, frameworks, and numerical libraries. An important facet of this work is devising common interfaces to encapsulate pre-existing physics modules, thereby facilitating code reuse among the computational accelerator team and fostering the incorporation of new capabilities developed over the lifetime of this project.
Such common interfaces will then help accelerator physicists to explore performance tradeoffs among different algorithms and implementations for different simulation scenarios and target petascale machines. For example, the parallel solution of large-scale collective effects calculations constitutes a significant fraction (typically 50-90 percent) of overall simulation time for Synergia, which is responsible for beam dynamics modeling in the COMPASS project. It is not clear a priori how to select and parameterize Synergia's current collective effects solver, an FFT-based Poisson solver, optimally for a given simulation instance. Moreover, efficiently calculating collective effects in the presence of large disparities in length scale could benefit from considering new algorithmic ideas and performance optimizations.
Our approach builds upon ongoing work on Synergia2 and MaryLie/IMPACT and will extend to other subdomains of the project. We are developing CCA-based infrastructure for component-based accelerator simulations, with a goal of facilitating the interaction among multiple physics modules, such as space charge, electron cloud, and wakefield effects, and providing easy access to computational tools under development by SciDAC2 projects, such as TOPS, ITAPS, VACETS, TASCS, and PERI. For example, through common interfaces, prototype TOPS solver components offer easy access to a range of (non)linear solvers developed by different groups at different institutions, including hypre (LLNL), PETSc (ANL), and SuperLU (LBNL).
Collaboration Notes: We have defined high-level components for beam dynamics based on Synergia. We are currently refactoring these interfaces to develop medium-grain components for particle tracking. We are also exploring the extension of prototype TOPS solver components in beam dynamics simulations.
Collaboration Image URLs: n/a
Software Involved: Synergia2, MaryLie/Impact, CCA toolkit, TOPS solver components, FFTW, PETSc, hypre, SuperLU
Collaboration URL: https://compass.fnal.gov/
Partner Project URL: https://compass.fnal.gov/
Partner Project Sponsorship: SciDAC Application/SAP
Sponsorship Details: SciDAC SAP: ASCR, SciDAC Application: HEP/NP/BES
