High performance computing (HPC) is a necessary component of modern astrophysics research. MKI has used 3 HPC clusters over the past decade. The first cluster has been used by the Chandra space telescope for data analysis and modeling associated with the HETG instrument contract. It was retired in October 2013.
The second cluster was used by the LIGO project, and was recently retired. As LIGO undergoes upgrades to Advanced LIGO sensitivity, the computing and analysis infrastructure will be completely overhauled.
The third cluster was originally built by Professors Edmund Bertschinger and Scott Hughes, and largely used by their research groups. It was significantly upgraded in 2008, replacing the original 32-bit compute nodes with 64-bit nodes. It has been further upgraded to use CentOS Enterprise Linux with SLURM workload management software. Slurm is an open source, fault-tolerant, and highly scalable cluster management and job scheduling system. The HPC cluster currently consists of 51 compute nodes and one master node, plus 4 storage servers providing researchers 1,048 CPU processors (cores) with 2,832 GB memory and over 1 PB data storage.
Over the past several years, this cluster has become a major computing resource for many MKI research groups. The following groups have used the cluster for a variety of MKI research projects:
Modeling of gravitational-wave sources and their measurement by gravitational-wave detectors. Modeling sources requires solving the equations of generic relativity to high accuracy, which requires large, parallel, high-precision HPC simulations. Modeling gravitational-wave measurement requires large Monte-Carlo simulations in order to explore the parameter space of signals that nature might provide and to understand how well these signals can be distinguished from one another.
Evolving with high precision and few approximations a system of N particles that interact through gravity.
Computing x-ray scattering from dust in the interstellar medium (ISM).
Computing exoplanet atmosphere models.
Computing models of galactic haloes, and simulating galaxy formation. Galaxy formation is a complicated nonlinear process. Researchers run large-scale N-body simulations to compare their theoretically predicted galaxies to observed ones. Because these programs are computationally expensive and require lots of memory, Frebel’s group uses the MKI HPC cluster to run and analyze these N-body simulations in order to gain intuition about the formation of galaxies.
Researching broadly cover structure and galaxy formation, dark matter physics and large-scale hydrodynamical simulations. His group makes extensive use of numerical simulations using state-of-the-art high-performance supercomputers around the world.