Radio astronomy beam forming on GPUs and the Cell processor

From Master Projects
Jump to: navigation, search

has title::Radio astronomy beam forming on GPUs and the Cell processor
status: finished
Student name: student name::Alessio Sclocco
Start start date:=2010/02/01
End end date:=2010/08/15
Supervisor: Rob van Nieuwpoort
Second reader: has second reader::Ana Lucia Varbanescu
Poster: has poster::Media:Media:Posternaam.pdf

Signature supervisor



A recent development in radio astronomy is to replace traditional radio telescope dishes with many small antennas. The signals are combined to form one large, virtual telescope. This process is both data and compute intensive. The LOFAR telescope, for instance, will produce over 100 terabytes per day. This data is currently processed on an IBM Blue Gene supercomputer with more than 10000 cores. The Blue Gene performs a data reduction step, and stores the data on disk.

An important processing step for the LOFAR telescope is "beam forming". This step selects a number of viewing directions that the instruments looks at in the sky. The beam forming algorithm used is not very difficult, but the challenge is to parallelize it efficiently, since it is very memory intensive. We have an implementation of this processing step on the Blue Gene supercomputer. However, we would like to have implementations on new architectures, like graphics processors (GPUs) from NVIDIA and ATI, and the Cell processor (from the playstation 3). These new devices exploit parallelism because they have many compute cores. These new architectures provide enormous compute power at very low cost. We have already ported several parts of the pipeline to these new processors, and they work well. We are especially interested in the new openCL language, which is portable across multi-core CPUs, the GPUs and the Cell.

So, this project requires the implementation of the beam forming algorithm on GPUs and the Cell processor. It will also give you the opportunity to work on a challenging and exciting real-world application.