Ericsson, Morgan: Toward an Optimizing Architecture of Scientific Computations using Parallel Components

Avhandlingen kommer att offentligt försvaras den 14/1, 2005, kl. 10.00 i sal D1136, Universitetsplatsen 1, Växjö universitet. Ansvarig institution är Matematiska och systemtekniska institutionen.

Abstract
This thesis presents an architecture that can select and compose a system of scientific components, and schedule it to computational nodes. The scheduling is continuous and strives to optimize the system according to a set of quality of service requirements. The thesis further presents a mapping of the architecture to Grid computing.

All scientific components are implemented using parallel programs. The system is a set of components and connections between these. The mapping of a system to a set of computational nodes is flexible.

The architecture presented is general enough to support distributed or parallel applications as long as they adhere to certain requirements. It is however primarily intended for the LOIS/LOFAR digital radio telescope, a set of sensor networks receiving data from space. These sensor networks produce an extreme amount of data at a speed of several gigabits per second. The data from different sensors must be merged in order for results to be obtained and several transformations, for instance corrections must be performed on the data stream.