BeschreibungFacing complex models in finance one has to consider to take advantage of parallel
computing. With the availability of multicore architectures even in commodity computers
there is an increased demand for practical strategies for utilizing these architectures.
Generally there are two different types of architectures: shared memory systems and distributed
memory systems. Each of which has its advantages and disadvantages which have
to be considered when creating parallel applications.
In this talk we present cluster@WU, a cluster of workstations employed at the Wirtschaftsuniversit
at Wien, and how it is utilized to solve computational problems. Furthermore,
we present strategies for parallelizing programs using different packages available in R. On
the basis of an example in numerical algebra we illustrate how both hardware architectures
can be used to achieve higher performance: For distributed memory systems such as clusters
of workstations we show how MPI can be used to explicitly parallelize a program. For
shared memory systems OpenMP can improve the performance of a sequential program
by implicit (compiler-driven) parallelization. Finally, we present results of a benchmark
experiment comparing the presented parallel routines with their sequential counterpart.
|28 Juni 2008 → 2 Juli 2008