Projects

The group has been successful in a number of scientific project grants relevant to the area of expertise, including the COST Action: ”Open European Network for High Performance Computing on Complex Environments” (2009-2013), a Faculty Award Research Grant from IBM: "Smart Cloud Infrastructures" (2009-2011), the GridCOMP STREP EU Project (2006-2010) and the CoreGRID NoE EU Project (2005-2009).

The group's recent and current projects include:

  • Portable implementation of the FFT benchmark
  • PMPI: high-level message passing in Fortran and C
  • JavaMPI: portable message passing in Java
  • Benchmarks in Java (ParkBench Low-level; NPB)
  • High Performance Distributed Processing with Replicated Jini Services

Software

PMPI: a Programmer's Message-Passing Interface

We have designed PMPI, a programmer's interface based on MPI that significantly simplifies application programming.

A prototype preprocessor for PMPI in both Fortran-77 and C has been developed.

The ParkBench single-processor low-level benchmarks in Java

  • POLY1: In-cache test. This benchmark tests severity of memory bottlenecks
  • POLY2: Out-of-cache test
  • RINF1: Arithmetic evaluation of r-infinity and n1/2 parameters
  • TICK1: Estimation of the clock tick resolution for system timing routines
  • TICK2: Test of the validity of the clock ticks-to-seconds multiplier

JavaMPI: a Java binding for MPI

This binding allows Java programs to use a native Message-Passing Interface library for running on distributed-memory machines. The binding was created from the C MPI binding with the aid of the Java-to-C Interface generating tool (JCI).

Java bindings for numerical libraries (BLAS, PBLAS, LAPACK, ScaLAPACK)

The bindings allow Java programs to access native high-performance numerical libraries, and were created with the aid of the Java-to-C Interface generating tool (JCI).