Mira (supercomputer)

Mira (supercomputer)
Argonne National Laboratory (Flickr) / U.S. Department of Energy · CC BY 2.0 · source
Name	Mira
Manufacturer	IBM
Location	Argonne National Laboratory
Country	United States
Introduced	2012
Decommissioned	2019
Type	Supercomputer
Architecture	Blue Gene/Q
Peak flops	8.59 petaflops
Storage	26 petabytes (file system)
Operating system	CNK (Compute Node Kernel)

Mira (supercomputer)

Mira was a Blue Gene/Q family supercomputer deployed at Argonne National Laboratory and operated by the Argonne Leadership Computing Facility for use by the United States Department of Energy and international collaborators. Commissioned in 2012 and retired in 2019, Mira served researchers from institutions such as University of Chicago, Oak Ridge National Laboratory, Lawrence Berkeley National Laboratory, Sandia National Laboratories, and Los Alamos National Laboratory to advance projects funded through programs like the Exascale Computing Project and initiatives by the Office of Science (United States Department of Energy). Mira contributed to studies involving partners including IBM, Cray Inc., Intel Corporation, NVIDIA, and research consortia such as the Energy Sciences Network.

Overview and history

Delivered by IBM as part of a series that included predecessors like Sequoia (supercomputer) and successors such as systems at Oak Ridge National Laboratory, Mira was sited in the Leadership Computing Facility at Argonne near Lemont, Illinois. The procurement and deployment involved contracts with U.S. Department of Energy programs and coordination with agencies including the National Science Foundation, and technical collaborations with universities like Massachusetts Institute of Technology, Stanford University, University of California, Berkeley, and Princeton University. Mira was installed to support petascale science after milestones achieved by systems such as Roadrunner, Jaguar (supercomputer), and Titan (supercomputer). Operational phases included allocations through the INCITE program and user support from centers like the National Energy Research Scientific Computing Center and the Oak Ridge Leadership Computing Facility.

Architecture and hardware

Mira used the Blue Gene/Q architecture developed by IBM Research with a 5D torus interconnect and compute nodes based on the PowerPC A2 processor core. The system consisted of racks assembled from midplanes and node cards similar to designs used in Blue Gene/L and Blue Gene/P, and components were manufactured by IBM Systems and Technology Group. Mira's topology and interconnect were engineered alongside advances from companies like Mellanox Technologies and leveraged fabric ideas explored in projects at Lawrence Livermore National Laboratory and Sandia National Laboratories. The system integrated high-density packaging and cooling approaches influenced by deployments at National Energy Research Scientific Computing Center and maintenance practices from Argonne National Laboratory facilities. Storage and I/O subsystems interfaced with parallel file systems used by projects at NERSC, and Mira's power envelope and floor layout were informed by precedents set by Blue Waters and interventions from U.S. Office of Science facility planners.

Performance and benchmarks

Mira achieved a Linpack performance of about 8.59 petaflops, placing it on the TOP500 list among contemporaries such as Titan (supercomputer), Sequoia (supercomputer), K computer, Piz Daint, and Fugaku in later years. Benchmarks included applications from the SPEC family and custom scaling studies by teams from University of Illinois Urbana–Champaign, California Institute of Technology, Columbia University, and Purdue University. Performance tuning involved collaborations with compiler and tools groups at IBM Research, GNU Project, Intel Corporation, and software efforts like OpenMP, MPI, HDF5, and NetCDF. Power efficiency metrics were compared against records set by systems such as Eurora and facilities at Riken, and Mira informed energy-performance trade-offs investigated by Lawrence Livermore National Laboratory and Sandia National Laboratories.

Applications and scientific contributions

Mira supported simulations in disciplines led by institutions like Argonne National Laboratory, University of Chicago, Northwestern University, University of Michigan, and Columbia University for problems including cosmology, climate modeling, materials science, combustion, and nuclear engineering. Notable projects included cosmological structure formation connected to teams at Kavli Institute for Cosmological Physics, climate studies in collaboration with NOAA and researchers from Princeton University, and materials research aligned with the Materials Project and groups at Brookhaven National Laboratory. Computational chemistry and biomolecular modeling on Mira involved researchers from Harvard University, Yale University, University of Texas at Austin, and pharmaceutical partnerships informed by work at Sandia National Laboratories. High-fidelity simulations contributed to publications in journals such as Science (journal), Nature (journal), Physical Review Letters, Proceedings of the National Academy of Sciences, and Journal of Computational Physics and fed into experimental programs at facilities like Argonne Tandem Linac Accelerator System and Advanced Photon Source.

Software and programming environment

The software stack on Mira was managed by teams from Argonne National Laboratory and IBM and included system software derived from the Compute Node Kernel lineage and toolchains maintained by the Open Source Initiative communities and projects like GNU Compiler Collection, IBM XL compilers, and performance tools from Intel Corporation. Programming models exploited MPI and mixed-mode paradigms blending OpenMP and custom tasking libraries developed with contributions from groups at Lawrence Berkeley National Laboratory, Sandia National Laboratories, and University of Illinois Urbana–Champaign. I/O and data management utilized libraries such as HDF5 and Parallel NetCDF with workflows coordinated by science gateways and portals used by collaborators at XSEDE and the Globus (service). Visualization and analysis pipelines interfaced with packages from Kitware, researchers at Argonne Leadership Computing Facility, and community efforts tied to ParaView and VisIt.

Decommissioning and legacy

Mira was officially retired and decommissioned in 2019, with transitions managed by Argonne National Laboratory and stakeholders including the U.S. Department of Energy and academic partners such as University of Chicago and Northwestern University. Lessons from Mira influenced procurement and design for successor systems funded under the Exascale Computing Project and hardware roadmaps advanced at IBM Research, Cray Inc., Intel Corporation, and NVIDIA. The scientific outputs, software artifacts, and workflow practices developed for Mira became part of repositories and community projects supported by DOE Office of Science programs, archives maintained by Argonne Leadership Computing Facility, and training curricula at institutions like Massachusetts Institute of Technology and University of Illinois Urbana–Champaign. Its legacy informed architectures of later platforms including deployments at Oak Ridge National Laboratory, Lawrence Livermore National Laboratory, and international centers such as RIKEN and CSCS (Swiss National Supercomputing Centre).

Category:Supercomputers Category:Argonne National Laboratory Category:IBM