The Science Data Processor team is designing imaging software for the SKA telescope, a collaborative effort spanning many countries across the globe. The longevity of the telescope stands in sharp contrast with the fast developments in computing hardware and leads to unique requirements for the software which will process exa-bytes of data per day. Domain Specific Languages (DSL) can deliver structure aimed at modifiability at the algorithmic and hardware dependent levels, and (other) DSL's can provide automatic re-optimization as computing hardware evolves. After a very brief overview of the project and problem we will discuss the data flow programming paradigms emerging for exa-scale and describe a few self-optimizing algorithmic constructions. We finish by describing how we are putting these elements together to prototype a design for the imaging software.
Peter Braam has held senior faculty positions in Mathematics and Computing at Oxford and Carnegie Mellon, prior to starting 5 startup companies (of which 4 were successful) as an entrepreneur. His best known project is the Lustre file system which powers the majority of high end HPC systems, a derivative of which is the ext4 system running on virtually all Linux systems. During the last few years, Peter has joined the research efforts at the Cavendish Laboratory in Cambridge, surrounding the SKA telescope and future parallel computing architectures. http://www.peterbraam.com