Speaker: Dr Michael Creel
Institution: Universitat Autònoma de Barcelona, Spain
The tutorial will show how a non-dedicated cluster can be created
in minutes. Upon shutdown, all computers in the cluster return to their
original state, so their temporary use in the cluster does not affect their
ordinary use. The cluster can be used for MPI-based parallel computing,
which can dramatically speed up computationally demanding calculations.
Illustrative examples include Monte Carlo simulation, econometric estimation
by GMM, and the Parameterized Expectations Algorithm for solving stochastic
dynamic macroeconomic models.
Organization
First session (1 hour): Introduction to clusters and MPI-based parallel
computing. Introduction to the ParallelKnoppix Linux distribution. Creation
of a cluster using ParallelKnoppix. A first brief example of the speedup
obtained through parallel computing.
Second session (1 hour): Examples of the use of MPI-based parallel computing
for problems in economics. Monte Carlo simulation; estimation by GMM; the
Parameterized Expectations Algorithm
Third session (1 hour): Writing parallel programs with GNU Octave as an
example. Advanced use and customization of the ParallelKnoppix CDROM.