UConn Physics Beowulf Cluster Group

(original proposal in pdf format)
 

1. Experimental Particle Physics (R.T. Jones, P.I.)

The cluster is used for Monte Carlo simulations of experiments and the analysis of large-scale experimental data sets. Monte Carlo simulations are parallelized simply by slicing the data into several equal-sized slices, and starting each slice from an independent pseudo-random sequence. Data analysis is likewise carried out in independent parallel streams. The i/o bottleneck is reduced by the use of the dCache Storage Resource Manager (SRM) which maintains the data across many disks spread over many nodes, providing the data files to the analysis program as if there were all on a single filesystem. The Parallel Root Facility (PROOF) cluster software from the CERN program library is used to handle parallel analysis of large sets of root files. The cluster has been configured as the University of Connecticut site of the Open Science Grid.

2. Theoretical Atomic/Molecular Physics (J. Javanainen, R. Cote, P.I.s)

The cluster is used for the numerical solution of theoretical models of many-atom systems. So far the codes have not been effectively parallelized, so the primary advantage afforded by the cluster for these studies is the possibility of running several long jobs at once.

3. Theoretical Condensed Matter Physics (G. Fernando, M. Rasamny, P.I.s)

The cluster is used for calculating the properties of materials starting from the basic electron-electron interactions in the solid. The calculations are carried out in the density functional formalism, and involve the manipulation of large arrays. The codes have been parallelized to allow the distribution of these arrays across many independent nodes, with the inter-process communication taking place using the Message-Passing Interface (MPI) library. A cluster-enabled version of the MPI library has been installed by the group and is actively used on the cluster. The group has customized the package to allow MPI to share the cluster with other non-MPI applications.