Workflows in Pulsar Astronomy.

In this chapter, we describe the development of methods that operate on the output of the signal of a radio telescope to detect the characteristic signals of pulsars. These signals are much weaker than the noise in the signal at any given wavelength, and therefore algorithms for combining the signals in different wavelength bands must be applied. This is heavily expensive in terms of CPU power. Early versions of distributed algorithms ran on a distributed network of supercomputers connected by Internet-aware Message Passing Interface (MPI) during the period 1999-2001. Today such techniques are being integrated into workflows that automate the search process and enable sophisticated astronomical knowledge to be captured via the construction of the workflow. In particular, we address issues of parallelism within components of the workflow. Parallelism is necessary due to two constraints on workflow performance. One is the application of the workflow in real time as the signal is being processed to enable very precise measurements to be carried out on known pulsars. The other is the use of the workflow to explore large regions of parameter space in search of previously undetected pulsars. There are very severe restraints on the degree of abstraction that can currently be applied in this work since details of the architecture of the computing resource (parallel cluster or computational Grid) on which the workflows are to be run cannot be ignored in the construction of the workflow. [ABSTRACT FROM AUTHOR]