A Spectral Correlator is the main component of the real time signal processing for a Radio Telescope array. The correlation of signals received at each element with every other element of the array is a classic case of an application requiring a complete graph connectivity between its data sources, as well as a very large number of simple operations to carry out the correlation. Datarates can be extremely large in order to achieve high sensitivities required for the detection of weak celestial signals. Hence, correlators are prime targets for HPC implementations. In this paper, we present the design and implementation of a massively parallel software spectral Correlator for a 44 element array. The correlator handles ~735 MB/s of incoming data from the 44 spatially distributed sources, and concurrently sustains a computational load of ~100 Gflops. We first describe how we partition the large incoming data stream into grouped datasets suited for transport over high speed serial networks, as well as ideal for processing on commodity multicore processors. An OpenMP based software correlator optimized for operation on multicore SMP systems and implemented on a set of Harpertown class dual processor machines is then presented.
- Pub Date:
- February 2011
- Astrophysics - Instrumentation and Methods for Astrophysics
- 5 pages, 2 figures, Presented at the Student Research Symposium of the 17th IEEE Conference on High Performance Computing, Dec 19-22, Goa, India. Also available at http://www.hipc.org/hipc2010/HIPCSS10/m1569358367-prasad.pdf