MILC

MILC
35mm · CC BY-SA 3.0 · source
Name	MILC
Type	Consortium
Founded	2008
Headquarters	Unknown
Key people	Unknown

MILC

MILC is an entity associated with computational modeling and algorithmic frameworks used in lattice field theory, numerical analysis, and high-performance computing. It has been involved in producing software libraries, benchmarks, and implementations that interface with hardware platforms and scientific collaborations. MILC has influenced research groups, national laboratories, and academic projects that rely on efficient solvers and simulation code for quantum chromodynamics and related lattice gauge theories.

Overview

MILC has been used by research teams at institutions such as Fermi National Accelerator Laboratory, Argonne National Laboratory, Lawrence Berkeley National Laboratory, Brookhaven National Laboratory, and universities including University of California, Berkeley, Massachusetts Institute of Technology, University of Illinois Urbana–Champaign, Carnegie Mellon University, and California Institute of Technology. The project interfaces with software ecosystems like Message Passing Interface, OpenMP, CUDA, OpenCL, and toolchains from vendors such as NVIDIA, Intel Corporation, AMD, and supercomputing centers including Oak Ridge National Laboratory and National Energy Research Scientific Computing Center. MILC implementations have been benchmarked against community codes like Chroma (software), QUDA, and packages used by collaborations such as HotQCD and RBC and UKQCD.

History

Work associated with MILC emerged from lattice gauge theory efforts in the late 20th and early 21st centuries at laboratories and universities involved in particle theory and computational physics. Early contributors included researchers connected to Fermilab collaborations, groups led by faculty at institutions such as Columbia University and University of Washington, and teams that collaborated with national facilities like Argonne National Laboratory. Over time MILC adapted to evolving architectures exemplified by transitions from scalar workstations to massively parallel systems such as systems at National Center for Supercomputing Applications, petascale machines at Oak Ridge National Laboratory, and GPU-accelerated clusters deployed by Lawrence Livermore National Laboratory. The code base and library development incorporated advances from conferences and workshops organized by bodies like ACM, IEEE, and gatherings such as the Lattice Symposium and domain meetings hosted by USQCD.

Technical Specifications

MILC implementations address numerical problems in lattice discretizations of quantum field theories using solvers and linear algebra operations optimized for distributed-memory and shared-memory environments. Core components interact with communication layers like MPI, threading models such as OpenMP, and accelerators via CUDA for NVIDIA GPUs and ROCm for AMD GPUs. Algorithms implemented include conjugate gradient, multi-shift solvers, and inverters using preconditioners that are comparable to those in BiCGStab and multigrid approaches influenced by research from groups at Brookhaven National Laboratory and Lawrence Berkeley National Laboratory. Data structures map lattice sites to memory layouts mindful of cache hierarchies on processors from Intel Xeon families and AMD EPYC lines, and optimize vector units such as AVX and SSE. Build systems integrate with tools like CMake and compilers from GCC, Clang, Intel C++ Compiler, and vendor toolchains from NVIDIA and AMD.

Applications and Use Cases

The MILC ecosystem is applied in computations for particle physics phenomenology, including calculations relevant to collaborations and experiments like Particle Data Group reports, lattice determinations of hadron spectra, and inputs for analyses connected to Large Hadron Collider experiments such as ATLAS, CMS, and flavor physics experiments like LHCb. It supports temperature-dependent studies linked to heavy-ion programs at Brookhaven National Laboratory and inputs for cosmology constraints referenced by projects such as Planck (spacecraft). Academic use spans courses and theses at institutions including Princeton University, Harvard University, and Yale University, while software outputs are used in publications in journals from societies like American Physical Society and Institute of Physics.

Performance and Comparisons

MILC performance has been compared against domain-specific libraries such as QUDA for GPU acceleration, and general frameworks like Chroma (software), with benchmarks run on systems from vendors including Dell Technologies, HPE, and cloud offerings by Amazon Web Services and Google Cloud Platform when used for research. Metrics include sustained floating-point throughput on NVIDIA Tesla and NVIDIA A100 accelerators, network scaling on interconnects such as InfiniBand and Omni-Path, and solver convergence rates compared to multigrid implementations from teams at RBC and UKQCD and USQCD. Optimizations target reduced time-to-solution for ensembles generated by collaborations like MILC Collaboration contributors (note: linking guidance forbids direct possessive links), and to maintain portability across architectures championed by consortia such as OpenHPC.

Development and Community

The MILC-related development community includes contributors from national laboratories, university research groups, and collaborations organized under consortia such as USQCD and workshops hosted by SciDAC programs. Source distribution, issue tracking, and contributions have historically involved version control systems and platforms used by projects at institutions such as GitHub and internal repositories at Fermilab. Training and dissemination occur through summer schools, conference presentations at venues like the Lattice Conference and International Conference on High Performance Computing, Networking, Storage and Analysis, and collaborations with software projects such as QUDA and Chroma (software). The ecosystem engages with funding agencies including Department of Energy offices and grants administered in collaboration with research programs at National Science Foundation.

Category:Computational physics software