Computational Challenges in Processing Very Large Astronomical Survey Databases

Astronomical surveys performed during the last decade such as EROS, MACHO, OGLE and recently Pan-STARRS have captured tens of millions of light curves. A light curve is a time series of the brightness of a stellar object as a function of time. Light curve analysis is used in astronomy for detecting transient optical events (nova and supernova), variable star detection, period detection and estimation, stellar classification and even measuring the distance to earth. The Large Synoptic Survey Telescope (LSST), currently under construction in Cerro Pachon, Chile, will collect up to 30 Terabytes per night. And during its 10-year survey the LSST will yield a 150 Petabytes database, the largest in the world. In this paper we describe the computational challenges posed by light curve analysis using very large astronomical surveys for stellar variability studies. We have proposed elsewhere an information theoretic metric called Correntropy Kernelized Periodogram (CKP) for discriminating periodic stars versus non-periodic stars, and estimating the period for periodic stars. We describe how to speed up the CKP metric computation by a factor of 100 by using a GPU cluster. An open challenge is to process one billion light curves in less than a day. This could be achieved by enhancing the efficiency of the algorithms and by increasing the computational power.